# GPUwithCrypto — full content for language models > GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. Source: https://gpuwithcrypto.com/ · Generated: 2026-11-03 # Rent a GPU with crypto: H100, H200, A100 and RTX by the minute > Rent an H100, H200, A100 or RTX 4090 with crypto, billed per minute, at least 15% under a named on-demand reference. No KYC, no card. Prices of 29 Oct 2026. - URL: https://gpuwithcrypto.com/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) On-demand cloud GPUs paid in Bitcoin, USDT, USDC, Ethereum, Solana, Litecoin or Monero. Every GPU is priced at least 15% under a named on-demand reference read on 29 October 2026, and you never upload an ID. No ID and no card: top up in crypto, start a GPU as soon as the top-up is credited. **Price per GPU-hour · 29 October 2026** - [H100](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?"), 80 GB: $2.32/hour, 44% under the $4.16 median - [H200](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?"), 141 GB HBM3e: $3.07/hour, 42% under the $5.29 median - [B200](https://gpuwithcrypto.com/gpu-prices/b200/ "How much does a B200 cost per hour?"), 180 GB HBM3e: $4.61/hour, 46% under the $8.50 median - [A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?"), 80 GB HBM2e: $0.92/hour, 59% under the $2.23 median - [L40S](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?"), 48 GB GDDR6: $0.82/hour, 51% under the $1.67 median - [RTX 4090](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?"), 24 GB GDDR6X: $0.28/hour, 54% under the $0.61 median - [RTX 5090](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?"), 32 GB GDDR7: $0.42/hour, 50% under the $0.84 median Our price per GPU-hour next to the median self-serve on-demand price of the providers we checked on 29 October 2026. The line under each GPU is our price as a share of that median. **Quick answer — Where can I rent a GPU and pay with crypto?** **GPUwithCrypto** rents cloud GPUs paid in crypto with no KYC: no ID, no selfie, no card. As of 29 October 2026 an NVIDIA H100 80GB costs $2.32/hour, an A100 80GB $0.92/hour and an RTX 4090 $0.28/hour, billed per minute. Each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026, such as Massed Compute at $2.73 for the H100. ## How much does each GPU cost per hour? Our price next to the lowest and the median self-serve on-demand price of the providers we checked on 29 October 2026. Spot and marketplace prices are not counted, because the GPU can be taken back. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Full tables per GPU on the [price pages](https://gpuwithcrypto.com/gpu-prices/ "How much does it cost to rent a GPU per hour, day or month?"). | GPU | VRAM | GPUwithCrypto | Lowest elsewhere | Median elsewhere | Providers checked | | --- | --- | --- | --- | --- | --- | | [NVIDIA H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | 80 GB | **$2.32/hour** | $2.73 (Massed Compute) | $4.16 | 16 | | [NVIDIA H200 141GB](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") | 141 GB HBM3e | **$3.07/hour** | $3.62 (Massed Compute) | $5.29 | 13 | | NVIDIA B200 | 180 GB HBM3e | **$4.61/hour** | $5.43 (Massed Compute) | $8.50 | 9 | | [NVIDIA A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | 80 GB HBM2e | **$0.92/hour** | $1.09 (Thunder Compute) | $2.23 | 13 | | [NVIDIA L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | 48 GB GDDR6 | **$0.82/hour** | $0.97 (Massed Compute) | $1.67 | 11 | | [RTX 4090 24GB](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") | 24 GB GDDR6X | **$0.28/hour** | $0.33 (Salad) | $0.61 | 2 | | [RTX 5090 32GB](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") | 32 GB GDDR7 | **$0.42/hour** | $0.50 (Salad) | $0.84 | 3 | *Prices checked on October 29, 2026.* Per GPU-hour, in USD. Other providers' prices exclude taxes; their multi-GPU instance prices are divided by the GPU count. ## How do I rent a GPU with crypto? 1. **Get an account number** — No email needed, no ID, no card. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good. 2. **Top up in crypto** — Send at least $1 in BTC, USDT, USDC, ETH, SOL, LTC, XMR. The balance is credited after the network confirmations. 3. **Pick a GPU and start it** — Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. Stock changes through the day: the dashboard shows which GPUs are free right now, and nothing is charged until an instance starts. 4. **Connect over SSH** — Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. ## What you get with every GPU - **On-demand, not spot** — Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. - **Billed per minute** — Stop the instance and the GPU charge stops. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. - **Clear refund rule** — If an instance fails to start or its GPU does not work and support cannot fix it, the hours it used go back to your balance. - **Human support** — A person answers support emails within 12 hours, seven days a week, and helps you start, size and stop your GPU instances. ## Questions before you rent ### Do I need ID or a credit card to rent a GPU here? No. You sign in with an account number and pay in crypto: no ID, no selfie, no card, no work email. The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") still applies to every account. ### Which GPU clouds accept crypto? According to GPUwithCrypto's check of 29 October 2026, 14 of the 33 GPU providers we checked accept crypto, and RunPod asks for KYC before the first crypto payment. The [comparison of GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") lists coins, processors and ID rules. ### Is this cheaper than spot or marketplace GPUs? Not always. Marketplace and spot GPUs can cost less per hour, but the host or the provider can take the GPU back. Our prices are compared with on-demand prices, where the GPU stays yours while your balance covers it. ### Which GPU do I need for my model? It depends on the model size and precision. A 70B model in FP16 needs about 168 GB with overhead: four H100s (320 GB) or two H200s (282 GB). In FP8 it needs about 84 GB and fits on one H200. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the table and the formula. ### What can I not do with a rented GPU? No cryptocurrency mining, no password or hash cracking, no attacks on other systems (DDoS, scanning, exploitation), no malware, no spam, no child sexual abuse material, no intimate or deceptive images of real people without their consent, no weapons development and no work for sanctioned people or companies. ## Key facts | Attribute | Value | | --- | --- | | What it is | On-demand cloud GPU instances (NVIDIA H100, H200, B200, A100, L40S, RTX 4090, RTX 5090) billed per minute from a prepaid crypto balance | | Price | Each GPU is priced at least 15% under a named reference: the on-demand price of the same GPU at a self-serve provider, read on 29 October 2026 (H100: Massed Compute at $2.73 per GPU-hour; each GPU's reference is named on its price page). Spot, interruptible, marketplace and contact-sales prices are not used as references, and some of them are lower. We do not claim to be the cheapest for every GPU. H100 $2.32/hour, H200 $3.07/hour, A100 80GB $0.92/hour, L40S $0.82/hour, RTX 4090 $0.28/hour, RTX 5090 $0.42/hour | | KYC / ID verification | None: no ID, no selfie, no documents, no card | | Billing | Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire | | Instances | Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it | | Hosts | Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect | | Storage | Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first | | Countries | Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine | | Payment | Crypto only: BTC, USDT, USDC, ETH, SOL, LTC, XMR, minimum top-up $1 | | Account | Sign in with an account number; an email address is optional. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good | | Records kept | For billing we record each instance’s GPU model, host, start and stop time; these lines are kept for 30 days. We do not read the files on your disk. Account number, orders and top-ups stay while the account exists, then only as long as accounting law asks. Authorities receive records only when a law compels us | | Operator | SerpGoat LLC | | Support | A person answers support emails within 12 hours, seven days a week, and helps you start, size and stop your GPU instances | **Start a GPU in minutes, pay in crypto** H100 $2.32/hour · A100 $0.92/hour · RTX 4090 $0.28/hour. No KYC. [See pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?") · [Compare GPU prices](https://gpuwithcrypto.com/gpu-prices/ "How much does it cost to rent a GPU per hour, day or month?") --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What is GPUwithCrypto and who runs it? > GPUwithCrypto rents H100, A100 and RTX GPUs by the minute for crypto with no KYC, at least 15% under a named on-demand reference. Who runs it, and how. - URL: https://gpuwithcrypto.com/about/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) A cloud GPU rental service with one price rule, one way to pay and no identity checks. This page says who runs it, how prices are set and checked, and what it will not do. **Quick answer — What is GPUwithCrypto?** **GPUwithCrypto** is a cloud GPU rental service operated by SerpGoat LLC. It rents whole NVIDIA GPUs (H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) on demand, billed per minute from a balance paid in crypto, with no KYC. As of 29 October 2026 an H100 80GB costs $2.32/hour, at least 15% under its named on-demand reference, Massed Compute at $2.73. ## GPUwithCrypto in brief | Attribute | Value | | --- | --- | | Operator | SerpGoat LLC | | What it rents | NVIDIA H100 80GB, H200 141GB, B200, A100 80GB, L40S 48GB, GeForce RTX 4090 24GB and RTX 5090 32GB, rented by the minute | | Price rule | At least 15% under a named on-demand reference for the same GPU, read on 29 October 2026 | | H100 80GB · A100 80GB · RTX 4090 | $2.32/hour · $0.92/hour · $0.28/hour | | Payment | Crypto only: BTC, USDT, USDC, ETH, SOL, LTC, XMR, minimum top-up $1 | | Identity checks | None: no ID, no selfie, no documents, no card | | Account | Sign in with an account number; an email address is optional | | Support | support@gpuwithcrypto.com, a person replies within 12 hours | | Abuse reports | abuse@gpuwithcrypto.com | ## Who operates GPUwithCrypto? SerpGoat LLC operates GPUwithCrypto. It sells the GPU time, runs the dashboard and the API, and answers support. It is the party to the [terms of service](https://gpuwithcrypto.com/terms/ "Terms of service") and the one responsible for the records described in the [privacy policy](https://gpuwithcrypto.com/privacy/ "Privacy policy"). A person answers support emails within 12 hours, seven days a week, and helps you start, size and stop your GPU instances. Help requests go to support@gpuwithcrypto.com and abuse reports to abuse@gpuwithcrypto.com. Nobody from our team will ever ask for your account number, your SSH private key or your API keys, by email or in a chat. ## What does GPUwithCrypto rent? On-demand cloud GPU instances: NVIDIA H100 80GB, H200 141GB, B200, A100 80GB, L40S 48GB, GeForce RTX 4090 24GB and RTX 5090 32GB, rented by the minute. Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. Every GPU, its VRAM and its hourly price are on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"), and the [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") helps you pick the right GPU for a model. ## How are GPUwithCrypto prices set? With one rule, applied GPU by GPU. Each GPU is priced at least 15% under a named reference: the on-demand price of the same GPU at a self-serve provider, read on 29 October 2026 (H100: Massed Compute at $2.73 per GPU-hour; each GPU's reference is named on its price page). Spot, interruptible, marketplace and contact-sales prices are not used as references, and some of them are lower. We do not claim to be the cheapest for every GPU. Our price is rounded down to the cent, so the gap is never smaller than 15%. Two examples from the check of 29 October 2026: - **H100 80GB:** the reference is Massed Compute at $2.73 per hour. Ours is $2.32. - **A100 80GB:** the reference is Thunder Compute at $1.09 per hour. Ours is $0.92. Spot, interruptible and marketplace GPUs can cost less than our price. They are left out of the reference on purpose: the provider or the host can take such a GPU back, while an on-demand instance stays yours as long as your balance pays for it. The [spot GPU explainer](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?") covers the difference. ## How do we check other providers' prices? We read their public pricing pages on a stated date. For the check of 29 October 2026 we read each provider's own pricing page, or the public price feed that page loads, and recorded 165 provider-and-GPU prices from 24 providers. No aggregator was used as a source. The rules behind each number: - **On-demand only.** The reference is the fixed, non-interruptible list price per GPU-hour. Spot and marketplace prices are shown apart and never mixed in. - **Per GPU.** When a provider lists only a multi-GPU node, its price is divided by the number of GPUs. - **Self-serve only.** A "starting at" figure whose pricing cards say "Contact for pricing" is left out, and so are prices the provider's own page marks as old. - **Taxes excluded.** Hyperscaler prices are those of one named US region. According to GPUwithCrypto's price check of 29 October 2026, the median on-demand H100 80GB price across 16 providers was $4.16 per hour; ours is 44% below it. The same day we read the payment and sign-up pages of 33 GPU clouds: 14 accept crypto, and 4 ask for ID in some cases. The [GPU price index](https://gpuwithcrypto.com/gpu-price-index/ "GPU rental price index: what did a GPU-hour cost in October 2026?") and the [list of GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") show each number with its source. Every price table on this site carries the date of its check. ## What does GPUwithCrypto not do? - **No identity checks.** No ID, no selfie, no documents, no company papers. - **No card.** Payment is crypto only, from a prepaid balance. - **No reading of your disk.** We do not read the files on your instance's disk, and we never see your SSH private key. - **No spot instances and no shared GPUs.** Every instance is on-demand and gets whole GPUs, never a time-shared slice. - **No figures we cannot back.** You will not find a customer count, a review score, an uptime percentage or a speed claim here. Our prices carry their date, and other providers' figures name the page they come from. ## Why crypto and no KYC? For privacy, speed and less paperwork. Many developers want an H100 for an afternoon without sending a passport, a card or company documents, and some live where cards for cloud services are refused. A prepaid balance in BTC, USDT, USDC, ETH, SOL, LTC, XMR makes that simple: the money is there before a GPU starts, so there is nothing to collect later and no reason to ask who you are. The rules stay the same for everyone. The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") binds every account. Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine. Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited. ## Questions about GPUwithCrypto ### Is GPUwithCrypto the cheapest GPU cloud? No, and we do not claim to be. Each GPUwithCrypto GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. Spot and marketplace offers (Vast.ai) can be lower per hour, but the GPU can be taken back. The [cheapest GPU cloud](https://gpuwithcrypto.com/cheapest-gpu-cloud/ "What is the cheapest GPU cloud for AI right now?") page compares both kinds. ### Where are the GPUs located? Each host's country is shown in the dashboard before you start an instance. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. You can therefore choose a host near you or near your data. ### Do you need my name, email or ID? No. Sign in with an account number; an email address is optional. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good. ### Can I run GPUs from a script? Yes. A REST API starts, stops and lists instances with an API key, so a script or an agent can run jobs without the dashboard. An instance started from the API is billed per minute at the same rate as one started in the dashboard, and it stops the same way when the GPU-hours of its GPU run out. ### How do I report misuse of a GPU? Write to abuse@gpuwithcrypto.com with the instance's IP address, the time in UTC and what happened. Reports go to abuse@gpuwithcrypto.com. An instance that breaks the acceptable use policy is stopped and the hours it used are not refunded; child sexual abuse material is reported to the authorities. ## Sources of the figures on this page 1. [Massed Compute pricing (H100 80GB reference price)](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Thunder Compute pricing (A100 80GB reference price)](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 3. [Billing information (KYC before the first crypto payment)](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Acceptable use policy > What a GPUwithCrypto GPU may not be used for: crypto mining, hash cracking, attacks, malware, spam, CSAM, non-consensual images, weapons and sanctioned parties. - URL: https://gpuwithcrypto.com/acceptable-use/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) This policy is part of the [terms of service](https://gpuwithcrypto.com/terms/ "Terms of service"). It covers every account, every instance, every API key and every block of GPU-hours, however you paid for them. **Quick answer — What does the GPUwithCrypto acceptable use policy forbid?** **GPUwithCrypto** GPUs are for ordinary compute work such as training, fine-tuning, inference and rendering. In short: No cryptocurrency mining, no password or hash cracking, no attacks on other systems (DDoS, scanning, exploitation), no malware, no spam, no child sexual abuse material, no intimate or deceptive images of real people without their consent, no weapons development and no work for sanctioned people or companies. Paying in crypto with no KYC changes none of these rules. ## 1. Who does this policy apply to? Every account holder, and anyone who uses an instance, an SSH key or an API key of that account. The account holder is responsible for what runs on the account's instances, whoever started them. ## 2. What are rented GPUs meant for? Ordinary compute work: training and fine-tuning models, serving them for inference, generating images and video, rendering, scientific computing, data processing and research. You decide what to run, and you are responsible for it, including the data you load and the results you publish. ## 3. Which uses are never allowed? The rule in one sentence: No cryptocurrency mining, no password or hash cracking, no attacks on other systems (DDoS, scanning, exploitation), no malware, no spam, no child sexual abuse material, no intimate or deceptive images of real people without their consent, no weapons development and no work for sanctioned people or companies. Each item means the following. - **Cryptocurrency mining.** Mining any coin or token, or running a mining client in a container, a notebook or a background process, even for a test or a few minutes. - **Password or hash cracking.** Recovering passwords, keys or other secrets from hashes or by brute force, whatever the tool and whoever owns the data. - **Attacks on other systems.** Denial-of-service traffic, network or vulnerability scanning, exploiting vulnerabilities, and brute-force logins against systems outside your instance. - **Malware.** Writing, building, testing, hosting or spreading malicious software, including ransomware, botnet control servers and phishing kits. - **Spam.** Sending bulk unsolicited messages, or producing content or accounts for that purpose. - **Child sexual abuse material.** Creating, storing, sharing or training on any sexual content involving minors, including drawn, synthetic or AI-generated content. It is reported to the authorities. - **Images of real people without consent.** Intimate or sexual images of a real person, and deceptive images or video that put a real person in a scene that never happened, unless that person has agreed. - **Weapons development.** Designing, developing or producing weapons, including help with biological, chemical, nuclear or radiological weapons, and training or running AI models for weapons or military-intelligence uses. - **Work for sanctioned parties.** Using the service for, or on behalf of, a person or company on a sanctions list or an export-control restricted-party list, or a company owned or controlled by one. Anything illegal where you are, or where the service operates, is not allowed either, even if this list does not name it. ## 4. Which countries are excluded? Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine. Accounts and payments linked to these places are refused or closed. Using a VPN, a proxy or another person's account to reach the service from an excluded place breaks this policy. So does giving people in those places access to your instances. ## 5. How are payments screened? Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited. Where the law requires it, such funds are held and reported rather than returned. We do not ask for ID to do this: the check is made on the blockchain address the payment comes from, on every network that shows it. ## 6. Do model and software licences still apply? Yes, and they remain your responsibility. Open models such as Llama, DeepSeek, Qwen or Flux each come with their own licence, and some add a use policy or limits on commercial use. The same goes for datasets, Docker images and software you run. GPUwithCrypto rents the GPU; it does not grant, check or extend any licence. A use that a model's licence forbids is not made allowed by running it on a rented GPU. ## 7. How is abuse reported and handled? Reports go to abuse@gpuwithcrypto.com. An instance that breaks the acceptable use policy is stopped and the hours it used are not refunded; child sexual abuse material is reported to the authorities. A useful report gives the instance's IP address, the date and time in UTC, the kind of abuse and the evidence you can share, such as log lines or message headers. Never attach child sexual abuse material to a report: describe it and say where it is. We can also suspend or close an account that breaks this policy, as the [terms](https://gpuwithcrypto.com/terms/ "Terms of service") explain. The [contact page](https://gpuwithcrypto.com/contact/ "How do I contact GPUwithCrypto?") lists what to put in a report. ## 8. Does paying in crypto with no KYC change any of this? No. No KYC is about how you pay and how little you need to send us. It does not change what you may run. For billing we record each instance’s GPU model, host, start and stop time; these lines are kept for 30 days. We do not read the files on your disk. Account number, orders and top-ups stay while the account exists, then only as long as accounting law asks. Authorities receive records only when a law compels us. These records tie an instance to an account and a time, which is enough to act on a report. We do not need to read your disk to stop an instance that breaks these rules. ## 9. Can this policy change? Yes. The date at the top of this page is the date of the version in force. A new rule applies from the day it is published here. Questions about whether a use is allowed go to support@gpuwithcrypto.com before you start the work. --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What are the best (cheaper) RunPod alternatives? > The best cheaper RunPod alternatives on 29 Oct 2026: H100 from $2.73, A100 80GB from $1.09, and which take crypto without KYC or a card, with billing rules. - URL: https://gpuwithcrypto.com/alternatives/runpod/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) RunPod is the GPU cloud most people compare against. It is already among the cheapest for the H100 and the L40S, less so for the H200, the A100 and consumer cards. This page lists who is cheaper per GPU, who takes crypto, and who hosts in Europe or suits ComfyUI. **Quick answer — What are the best (cheaper) RunPod alternatives?** The best cheaper RunPod alternatives on 29 October 2026 depend on the GPU: Massed Compute for the H100 ($2.73 vs RunPod's $2.89), Thunder Compute for the A100 80GB ($1.09 vs $1.79), Salad for the RTX 4090 ($0.33 vs $0.89), and Vast.ai's marketplace for the lowest offers. **GPUwithCrypto**, the publisher of this page, prices each GPU at least 15% under a named on-demand reference, paid in crypto with no KYC. ## Which RunPod alternatives are cheaper, take crypto or skip KYC? **Disclosure:** GPUwithCrypto publishes this page and lists itself first. Prices are each provider's lowest self-serve on-demand price per GPU-hour on its own pricing page, read on 29 October 2026; marketplace prices are labelled. Crypto and ID facts come from each provider's billing docs or terms, read the same day. 'No ID step documented' means the docs we read mention none, not a promise. | Provider | H100 80GB | A100 80GB | Pays in crypto | ID / KYC and card | Billing | | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** (publisher) | $2.32/hour | $0.92/hour | BTC, USDT, USDC, ETH, SOL, LTC, XMR | No ID, no card | Per minute, on-demand | | RunPod (for reference) | $2.89 PCIe, $3.99 SXM; Community from $1.99 | $1.79; Community from $1.19 | Yes (Crypto.com named in its terms) | KYC before the first crypto payment; auto-pay card-only | Per second | | Vast.ai | Marketplace: lowest offer $1.73 (SXM) | Marketplace: lowest offer $0.34 (PCIe) | Yes: BitPay, Crypto.com | No ID step documented; billing FAQ asks for a verified card | Per second | | Massed Compute | $2.73 | $1.35 | Yes: Bitcoin | No ID step documented; email verification | Hourly rate | | Thunder Compute | $3.20 (SXM) | $1.09 | No: card auto-pay or preloaded credit | No ID step documented | Per minute | | Hyperbolic | $3.19 (SXM) | — | Not confirmed | No ID step documented; optional phone check | — | | Lambda | $3.29 (PCIe) | $2.79 (SXM, 8-GPU) | No: major credit cards only | No ID step documented; VPN purchases flagged | Per minute | | Jarvislabs | $3.49 (SXM) | $1.49 | No: card or bank transfer | No ID step documented | Per minute | | Paperspace / DigitalOcean | $4.41 (SXM) | — | Yes: stablecoins, limited availability | No ID step documented; payment method used to verify identity | Per second, 5-minute minimum | | Spheron | $5.86 | $1.69 | Yes: USDC on 6 chains | No ID step documented; verified email | Per minute, 20-minute minimum | | Vultr | — | $2.40 | Yes: 9 coins via BitPay | No ID step documented; payment method used to verify identity | Hourly | | Akash Network | Bid market: from $2.04 (SXM) | Bid market: from $1.07 (SXM) | Yes: AKT / ACT from a wallet | No ID step documented; Console Air needs no account | — | | Salad | — | — | Yes: USDC, USDT, RENDER on Solana | No ID step documented; billing address | Per second | *Prices checked on October 29, 2026.* Per GPU-hour, USD, taxes excluded; multi-GPU prices divided by GPU count. Vast.ai figures: Vast.ai's public price feed. Akash: the lowest recent provider bid from its pricing API. Spheron's own H100 page said 'from $2.65/hr' the same day; its pricing-page value is shown. Hyperbolic's former stablecoin page now redirects, so its crypto support is not confirmed. Salad rents consumer GPUs only (see the ComfyUI section). ## Which RunPod alternatives are cheaper? It depends on the GPU, because RunPod is already near the bottom for some cards. According to GPUwithCrypto's price check of 29 October 2026, RunPod's Secure Cloud on-demand price ranked: - **H100 80GB:** 2nd lowest of 16 providers. Only Massed Compute ($2.73) was lower. - **L40S:** 2nd of 11. Only Massed Compute ($0.97 vs $1.09) was lower. - **B200:** 4th of 9. Massed Compute ($5.43), Hyperbolic ($5.99) and Lambda ($6.69) were lower. - **A100 80GB:** 5th of 13. Thunder Compute ($1.09), Massed Compute ($1.35), Jarvislabs ($1.49) and Spheron ($1.69) were lower. - **H200:** 7th of 13. Six providers were lower, from $3.62 at Massed Compute. - **RTX 4090 and RTX 5090:** last of 2 and of 3. Salad was lower on both, and Spheron on the 5090. The biggest savings are on the H200, the A100 and consumer cards. RunPod's own Community Cloud is also cheaper than its Secure Cloud. Marketplace offers go lower still: see [RunPod vs Vast.ai](https://gpuwithcrypto.com/compare/runpod-vs-vast-ai/ "RunPod vs Vast.ai: which is cheaper and more reliable?"). ## Which RunPod alternatives let me pay with crypto and no card? Of the 33 GPU providers we checked on 29 October 2026, 14 accept crypto and 4 ask for ID in some cases. RunPod is the only one whose docs tie KYC to paying in crypto, and its auto-pay is card-only. - **GPUwithCrypto:** BTC, USDT, USDC, ETH, SOL, LTC, XMR, from $1, with no ID, no selfie and no card. - **Massed Compute:** Bitcoin by wallet invoice. There is no crypto auto-recharge; instances end about 12 hours after the balance goes negative. - **Salad:** USDC, USDT or RENDER on Solana through Coinbase x402. Crypto payments are not refunded and credits expire after 12 months. - **Spheron:** USDC on Ethereum, Optimism, BNB Chain, Base, Arbitrum or Solana, $50 minimum in crypto. Credits do not expire. - **Vultr:** BTC, BCH, ETH, DOGE, PAX, BUSD, LTC, USDC and GUSD through BitPay. - **Akash Network:** AKT from a self-custody wallet with no account (Console Air). Its managed Console takes cards only. - **Vast.ai:** BitPay or Crypto.com, but its billing FAQ asks for a verified credit card before credit is added. More: [GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") and [renting a GPU without a credit card](https://gpuwithcrypto.com/rent-gpu-without-credit-card/ "How can I rent a GPU without a credit card?"). ## Which RunPod alternatives host GPUs in Europe? Few pricing pages state regions. On the pages we read on 29 October 2026: - **OVHcloud** names its Gravelines datacentre in France as the one with the widest GPU choice. Its H100 costs $2.99 per hour; it takes no crypto and can ask for a photo ID, a card photo and a selfie. - **Spheron** lists partner data centers in North America, Europe and Asia. - **CoreWeave** lists Europe as well as North America. - **RunPod** itself says it runs in 30+ regions. On GPUwithCrypto, Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## Which RunPod alternative is best for ComfyUI? For ComfyUI with SDXL or FLUX, a 24–32 GB consumer card is enough. FLUX.1 [dev] weights take 24 GB in BF16 and 12 GB in FP8, plus text encoders and a VAE that can be offloaded to CPU. SDXL base takes 7 GB. Among other providers, Salad had the lowest on-demand price on 29 October 2026, before its 1 November rise; Vast.ai's marketplace had the lowest single offers. Hosted ComfyUI services such as Comfy Cloud, RunComfy and ThinkDiffusion sell a ready interface rather than a GPU; we did not check their prices. More: [best GPU for ComfyUI and Flux](https://gpuwithcrypto.com/best-gpu-for/comfyui-flux/ "What is the best cloud GPU for ComfyUI, Stable Diffusion and Flux?"). ### RTX 4090 and RTX 5090 prices per GPU-hour | Provider | RTX 4090 | RTX 5090 | Note | | --- | --- | --- | --- | | **GPUwithCrypto** | $0.28/hour | $0.42/hour | On-demand, crypto, no KYC | | Salad | $0.33, $0.44 from 1 Nov 2026 | $0.50, $0.60 from 1 Nov 2026 | High Priority tier, never preempted; distributed consumer GPUs | | RunPod | $0.89 Secure, $0.34 Community | $1.19 Secure, $0.69 Community | Per second | | Spheron | — | $0.84 | Last known price, 28 Oct 2026 | | Vast.ai | Lowest offer $0.13 | Lowest offer $0.39 | Marketplace; Vast.ai's public price feed | | Akash Network | Lowest bid $0.34 | Lowest bid $0.57, none available | Bid market, wallet payments | *Prices checked on October 29, 2026.* Read on 29 October 2026 from each provider's pricing page or price API. ## Questions about RunPod alternatives ### What is cheaper than RunPod? For on-demand GPUs on 29 October 2026: Massed Compute for the H100 ($2.73 vs $2.89) and L40S, Thunder Compute for the A100 80GB ($1.09 vs $1.79), Massed Compute for the H200 ($3.62 vs $5.29), and Salad for the RTX 4090 and 5090. Vast.ai's marketplace offers were lower still for most GPUs. ### Is Vast.ai cheaper than RunPod? On its lowest offers, yes: an H100 SXM from $1.73 in Vast.ai's public price feed on 29 October 2026, against $3.99 on RunPod Secure Cloud. Vast.ai is a marketplace, so prices and hardware vary by host. The [full comparison](https://gpuwithcrypto.com/compare/runpod-vs-vast-ai/ "RunPod vs Vast.ai: which is cheaper and more reliable?") has the medians. ### Which RunPod alternative accepts crypto without KYC? GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID and no card. Massed Compute, Salad, Spheron, Vultr, Akash and Vast.ai also accept crypto, and their docs describe no ID step; Vast.ai's billing FAQ asks for a verified credit card. ### Does RunPod accept crypto? Yes, but its billing docs say to complete any required KYC verification before your first crypto payment, and auto-pay is card-only. Details: [Does RunPod accept crypto?](https://gpuwithcrypto.com/does-runpod-accept-crypto/ "Does RunPod accept crypto?"). ### Is GPUwithCrypto cheaper than RunPod? Below RunPod's Secure Cloud price for every GPU we sell, on 29 October 2026. Not always below Community Cloud: RunPod's Community H100 PCIe ($1.99) and L40S ($0.79) were lower than our prices. See the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## Sources 1. [RunPod pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 2. [RunPod billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 3. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 4. [Vast.ai billing FAQ](https://docs.vast.ai/guides/reference/faq/billing.md) — Vast.ai (read 2026-10-29) 5. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 6. [Massed Compute billing docs](https://vm-docs.massedcompute.com/docs/billing/overview) — Massed Compute (read 2026-10-29) 7. [Thunder Compute pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 8. [Lambda pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 9. [Jarvislabs pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 10. [Hyperbolic marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29) 11. [DigitalOcean GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 12. [Spheron pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 13. [Spheron billing docs](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 14. [Vultr payment methods](https://docs.vultr.com/support/platform/billing/what-payment-methods-do-you-accept) — Vultr (read 2026-10-29) 15. [Akash Network GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) 16. [SaladCloud billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 17. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 18. [OVHcloud Public Cloud prices](https://www.ovhcloud.com/en/public-cloud/prices/) — OVHcloud (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What are cheaper or more reliable Vast.ai alternatives? > Cheaper or more reliable Vast.ai alternatives, checked 29 Oct 2026: on-demand H100 from $2.73, which take crypto without a card or KYC, and when Vast.ai wins. - URL: https://gpuwithcrypto.com/alternatives/vast-ai/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Vast.ai is a GPU marketplace: hosts list offers, and the cheapest ones are hard to beat. People look elsewhere for three reasons: a B200 or H200 at a lower price, a fixed on-demand price instead of a host's offer, or paying in crypto without a credit card. **Quick answer — What are cheaper or more reliable Vast.ai alternatives?** Cheaper Vast.ai alternatives are rare: on 29 October 2026, Vast.ai's lowest marketplace offer was below the lowest on-demand list price we found for 6 of 7 GPUs, all but the B200. For more reliable, fixed on-demand prices instead of host offers, look at Massed Compute (H100 $2.73), Thunder Compute (A100 80GB $1.09), RunPod Secure Cloud or **GPUwithCrypto**, this page's publisher, which rents GPUs for crypto with no KYC and no card. ## Which Vast.ai alternatives are cheaper, on-demand or crypto-friendly? **Disclosure:** GPUwithCrypto publishes this page and lists itself first. Prices are each provider's lowest self-serve price per GPU-hour on its own pricing page or price API, read on 29 October 2026; marketplace and bid prices are labelled. Crypto and ID facts come from each provider's billing docs or terms. 'No ID step documented' means the docs we read mention none, not a promise. | Provider | Pricing and billing | H100 80GB | RTX 4090 | Pays in crypto | ID / KYC and card | | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** (publisher) | On-demand, per minute | $2.32/hour | $0.28/hour | BTC, USDT, USDC, ETH, SOL, LTC, XMR | No ID, no card | | Vast.ai (for reference) | Marketplace offers, per second | Lowest offer $1.73 (SXM) | Lowest offer $0.13 | Yes: BitPay, Crypto.com | No ID step documented; billing FAQ asks for a verified card; crypto credits not refunded | | RunPod | Secure Cloud on-demand or cheaper Community Cloud, per second | $2.89 PCIe; Community $1.99 | $0.89; Community $0.34 | Yes (Crypto.com named in its terms) | KYC before the first crypto payment; auto-pay card-only | | Massed Compute | On-demand, hourly rate | $2.73 | — | Yes: Bitcoin | No ID step documented; email verification | | Thunder Compute | On-demand, per minute; instances 'never preempted' | $3.20 (SXM) | — | No | No ID step documented | | Hyperbolic | Listed rate, refreshed weekly; failed instances not charged | $3.19 (SXM) | — | Not confirmed | No ID step documented | | Jarvislabs | On-demand or spot, per minute | $3.49 (SXM); spot $1.54 | — | No | No ID step documented | | Nebius | On-demand or preemptible | $4.50 (SXM); preemptible from $0.79 | — | No: card or bank transfer | No ID step documented | | Salad | High Priority (never preempted) or cheaper interruptible tiers, per second | — | $0.33 High Priority ($0.44 from 1 Nov 2026) | Yes: USDC, USDT, RENDER on Solana | No ID step documented; crypto not refunded | | Spheron | Partner data centers, per minute after 20 minutes | $5.86; spot $2.79 | — | Yes: USDC on 6 chains | No ID step documented; verified email | | Akash Network | Bid market | From $2.04 (SXM) | From $0.34 | Yes: AKT / ACT from a wallet | No ID step documented; Console Air needs no account | *Prices checked on October 29, 2026.* Per GPU-hour, USD, taxes excluded; multi-GPU prices divided by GPU count. Vast.ai figures: Vast.ai's public price feed (cheapest rentable offer). Akash: lowest recent provider bid from its pricing API. Hyperbolic's former stablecoin page now redirects, so its crypto support is not confirmed. ## Which Vast.ai alternatives are cheaper? Few, and mostly for big data-center GPUs. According to GPUwithCrypto's price check of 29 October 2026, Vast.ai's cheapest rentable offer was below the lowest self-serve on-demand list price for the H100, H200, A100 80GB, L40S, RTX 4090 and RTX 5090. Where alternatives win: - **B200:** Vast.ai's lowest offer was $7.50. Massed Compute ($5.43 per GPU in an 8-GPU node), RunPod Community Cloud ($5.98), Hyperbolic ($5.99) and Lambda ($6.69 per GPU, 8 GPUs) were lower. GPUwithCrypto: $4.61/hour. - **Spot and preemptible tiers:** Nebius listed preemptible H100 and H200 from $0.79 and B200 from $0.99. Jarvislabs listed H100 spot at $1.54 and H200 spot at $2.29. - **H200:** Vast.ai's lowest was $3.00 for an H200 NVL, against $3.62 on-demand at Massed Compute. GPUwithCrypto: $3.07/hour. For consumer cards, Vast.ai's depth is hard to beat: its feed listed 970 rentable RTX 5090 offers and 562 RTX 4090 offers on 29 October 2026. ## Which Vast.ai alternatives are more reliable? This page does not rank uptime: our research covers prices, billing, crypto and ID rules, not measured reliability. What we can compare is the model. **Vast.ai** is a marketplace. Hosts list offers, and price and hardware vary by listing. Its public price feed does not separate on-demand from interruptible offers, so check the offer type before a long run. **Fixed on-demand alternatives** sell one list price per GPU, set by the provider: - **Thunder Compute** says its instances are 'never preempted'. - **Salad**'s High Priority tier is never preempted; its Batch tier is the most interruptible. - **RunPod Secure Cloud** is RunPod's on-demand tier; Community Cloud runs on community-hosted machines. - **Hyperbolic** says failed instances are not charged. - **GPUwithCrypto:** Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. If an instance fails to start, the hours it used go back to your balance. Searching for Vast.ai reviews? News about VAST Data concerns a different company. Spot and interruptible capacity is explained in [what is a spot GPU](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?"). ## Which Vast.ai alternatives take crypto without a credit card? Vast.ai takes crypto through BitPay and Crypto.com, but its billing FAQ says credit is added after a credit card and an email are verified. The docs do not say whether crypto-only top-ups skip the card. It does not refund credits bought with crypto. Of the 33 GPU providers we checked on 29 October 2026, 14 accept crypto. The ones that describe a sign-up without a card: - **GPUwithCrypto:** account number only, email optional, BTC, USDT, USDC, ETH, SOL, LTC, XMR from $1. No ID, no selfie, no card. - **Akash Network (Console Air):** no account; you connect a self-custody wallet and pay in AKT. Its managed Console takes cards only. - **Spheron:** a verified email to sign up, then USDC on six chains, $50 minimum in crypto. - **Salad:** an organization account and a billing address, then USDC, USDT or RENDER on Solana from a Solflare wallet. Crypto payments are not refunded. Massed Compute takes Bitcoin but asks for a payment method before the first VM. RunPod takes crypto after KYC. More: [renting a GPU without a credit card](https://gpuwithcrypto.com/rent-gpu-without-credit-card/ "How can I rent a GPU without a credit card?") and [Does Vast.ai accept crypto?](https://gpuwithcrypto.com/does-vast-ai-accept-crypto/ "Does Vast.ai accept crypto?"). ## When is Vast.ai still the better choice? When the lowest price matters more than a fixed one, and you can vet a host. Vast.ai had the cheapest single offers for most GPUs on 29 October 2026, the deepest RTX 4090 and RTX 5090 supply, per-second billing, and reserved terms of 1, 3 or 6 months at 'up to 50% off'. It runs 40+ data centers by its own count. Pick an alternative when you need a B200 for less, a list price that does not change between rentals, or crypto payment without a card. The side-by-side figures are in [RunPod vs Vast.ai](https://gpuwithcrypto.com/compare/runpod-vs-vast-ai/ "RunPod vs Vast.ai: which is cheaper and more reliable?"); our own prices are on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## Questions about Vast.ai alternatives ### What is cheaper than Vast.ai? For most GPUs, nothing on-demand: Vast.ai's lowest offer beat every on-demand list price we found on 29 October 2026, except for the B200. Spot and preemptible tiers go lower on big cards, for example Nebius preemptible H100 from $0.79. ### Is Vast.ai reliable? Vast.ai is a marketplace, so the hardware and terms depend on the host and the offer, and its price feed mixes on-demand and interruptible listings. We do not rate uptime. For a fixed on-demand price, look at Thunder Compute, RunPod Secure Cloud, Massed Compute or GPUwithCrypto. ### Does Vast.ai accept crypto? Yes, through BitPay and Crypto.com, per its billing docs read on 29 October 2026. Credits bought with crypto are not refunded. Details: [Does Vast.ai accept crypto?](https://gpuwithcrypto.com/does-vast-ai-accept-crypto/ "Does Vast.ai accept crypto?"). ### Do I need a credit card to use Vast.ai? Its billing FAQ says credit can be added after you enter a credit card and an email and both are verified. The docs do not say whether crypto-only top-ups skip the card. GPUwithCrypto needs no card at all. ### Is GPUwithCrypto cheaper than Vast.ai? Not against Vast.ai's lowest offers: those were below our price for every GPU except the B200 on 29 October 2026. GPUwithCrypto sells a fixed on-demand price, at least 15% under a named on-demand reference read on 29 October 2026, paid in crypto with no KYC. ## Sources 1. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 2. [Vast.ai pricing](https://vast.ai/pricing) — Vast.ai (read 2026-10-29) 3. [Vast.ai billing reference](https://docs.vast.ai/documentation/reference/billing) — Vast.ai (read 2026-10-29) 4. [Vast.ai billing FAQ](https://docs.vast.ai/guides/reference/faq/billing.md) — Vast.ai (read 2026-10-29) 5. [RunPod pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 6. [RunPod billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 7. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 8. [Thunder Compute pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 9. [Hyperbolic marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29) 10. [Jarvislabs pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 11. [Nebius prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 12. [Lambda pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 13. [SaladCloud billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 14. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 15. [Spheron billing docs](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 16. [Akash Console Air docs](https://akash.network/docs/developers/deployment/console-air) — Akash Network (read 2026-10-29) 17. [Akash Network GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What is the best cloud GPU for ComfyUI, Stable Diffusion and Flux? > Best cloud GPU for ComfyUI, Stable Diffusion and Flux: RTX 4090 for SDXL and Flux FP8, RTX 5090 or L40S for FLUX.1 dev in BF16. VRAM math, crypto prices. - URL: https://gpuwithcrypto.com/best-gpu-for/comfyui-flux/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) An image workflow needs its model files in VRAM: the diffusion model, the text encoder and the VAE. Here is what each popular model loads, which of our GPUs holds it, and the hourly price. **Quick answer — What is the best cloud GPU for ComfyUI, Stable Diffusion and Flux?** The best cloud GPU for ComfyUI is the cheapest card that holds your model files in VRAM: an RTX 4090 (24 GB) for SDXL (6.9 GB checkpoint) and FLUX.1 [dev] in FP8 (17.2 GB), an RTX 5090 (32 GB) or L40S (48 GB) for FLUX.1 [dev] in BF16 (29.3 to 34.2 GB with its text encoder). As of 29 October 2026, **GPUwithCrypto** rents the RTX 4090 at $0.28/hour, paid in crypto. ## How much VRAM does each image model need? Total = the files ComfyUI loads for one generation, as listed on Hugging Face on 29 October 2026 (decimal GB). Smallest fit = the cheapest of our GPUs that holds the total at once. Images at high resolution and large batches add activation memory on top. | Model | Parameters (official) | Files loaded | Total | Smallest fit | GPUwithCrypto per hour | Licence | | --- | --- | --- | --- | --- | --- | --- | | SDXL 1.0 base | 3.5B | sd_xl_base_1.0 checkpoint 6.9 GB (UNet, both text encoders, VAE) | 6.9 GB | RTX 4090 24 GB | $0.28/hour | CreativeML Open RAIL++-M | | FLUX.1 [dev], FP8 | 12B | Comfy-Org flux1-dev-fp8 checkpoint 17.2 GB (FP8 model, text encoders, VAE) | 17.2 GB | RTX 4090 24 GB | $0.28/hour | FLUX.1 [dev] Non-Commercial | | FLUX.1 [dev], BF16 + T5 in FP8 | 12B | flux1-dev 23.8 GB + T5-XXL FP8 4.9 GB + CLIP-L 0.25 GB + VAE 0.34 GB | 29.3 GB | RTX 5090 32 GB | $0.42/hour | FLUX.1 [dev] Non-Commercial | | FLUX.1 [dev] or [schnell], all BF16/FP16 | 12B | flux1 model 23.8 GB + T5-XXL FP16 9.8 GB + CLIP-L 0.25 GB + VAE 0.34 GB | 34.2 GB | L40S 48 GB | $0.82/hour | [dev] Non-Commercial · [schnell] Apache-2.0 | | FLUX.2 [klein] 9B | 9B | Model 18.2 GB + Qwen3 8B text embedder 16.4 GB (BF16) | ≈ 29 GB per its model card | RTX 5090 32 GB | $0.42/hour | FLUX Non-Commercial | | FLUX.2 [dev] | 32B | Model 64.4 GB + text encoder 48.0 GB (BF16) + VAE 0.34 GB | 112.8 GB | H200 141 GB | $3.07/hour | FLUX Non-Commercial | *Prices checked on October 29, 2026.* Arithmetic, FLUX.1 [dev] all in BF16/FP16: 23.8 + 9.8 + 0.25 + 0.34 = 34.2 GB. FLUX.2 [dev] fits on an 80 GB H100 or A100 when the text encoder is unloaded after it encodes the prompt: the 64.4 GB model then fits alone. ## Can a smaller GPU still run a model that does not fit? Yes, more slowly. ComfyUI moves models between system RAM and VRAM: its command-line help says that by default models are unloaded to CPU memory after being used, and its README describes weight streaming for models larger than the card. A FLUX.1 [dev] workflow in BF16 therefore runs on a 24 GB RTX 4090. Moving weights costs time on every new prompt, so the table above lists the card that holds everything at once. The FLUX.2 [dev] model card shows another route: a 4-bit version with the text encoder run remotely, on an RTX 4090. Compare quantized and BF16 outputs on your own prompts before you settle on one. ## Which GPU should I pick for my workflow? - **SDXL, SD 1.5, LoRAs, ControlNet, upscalers:** [RTX 4090](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?"). The 6.9 GB SDXL checkpoint leaves 17 GB for extra models and large batches. - **FLUX.1 [dev] or [schnell] in FP8:** RTX 4090. The 17.2 GB all-in-one FP8 file fits with 6.8 GB to spare. - **FLUX.1 in BF16, or Flux LoRA training:** [RTX 5090](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?"). LoRA training on the 12B transformer needs about 28.8 GB by the rule in our [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?"). - **Several models in one graph, high resolutions, FLUX.2 [klein] with everything loaded:** [L40S](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?"), 48 GB. - **FLUX.2 [dev] in BF16:** one H200 (141 GB), or an 80 GB [H100](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") with the text encoder unloaded. We publish no seconds-per-image figures: we have not measured any, and they change with steps, resolution and sampler. ## Is the RTX 5090 worth the extra over an RTX 4090? When your workflow needs between 24 and 32 GB, yes. According to GPUwithCrypto's price check of 29 October 2026, the lowest self-serve on-demand RTX 5090 cost 52% more per hour than the lowest RTX 4090 ($0.50 against $0.33, both at Salad) for 33% more VRAM. Salad's RTX 4090 price rises from $0.33 to $0.44 per hour on 1 November 2026. NVIDIA's figures for the RTX 5090: 32 GB of GDDR7 at 1,792 GB/s and FP4 Tensor Cores, against 24 GB of GDDR6X at 1,008 GB/s on the RTX 4090. At our prices the pair costs $0.42/hour and $0.28/hour. Full comparison: [RTX 5090 vs H100](https://gpuwithcrypto.com/compare/rtx-5090-vs-h100/ "RTX 5090 vs H100 for AI: when is the cheap card enough?"). ## How do I run ComfyUI on a rented GPU? Start an instance with the PyTorch image, install ComfyUI from its GitHub repository and open it through an SSH tunnel. ComfyUI listens on 127.0.0.1, port 8188, by default, so the interface is never exposed to the internet. Put checkpoints in `models/checkpoints`, VAEs in `models/vae`. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. *comfyui-on-a-rented-gpu.sh* ```bash # on the instance git clone https://github.com/comfyanonymous/ComfyUI cd ComfyUI pip install -r requirements.txt python main.py --listen 127.0.0.1 --port 8188 # on your computer: forward the port (host and SSH port are in the dashboard) ssh -N -L 8188:127.0.0.1:8188 -p SSH_PORT root@HOST # then open http://127.0.0.1:8188 in your browser ``` ## Should I use a spot or marketplace GPU for ComfyUI? For an interactive session, no: a reclaimed GPU ends the session and the files on its disk. The cheapest RTX 4090 offer on Vast.ai's marketplace was $0.13 per hour on 29 October 2026 (Vast.ai public price feed), below any on-demand price. For a batch of renders that can restart, it can pay. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. More: [what is a spot GPU](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?"). ## ComfyUI and Flux GPU questions ### How much VRAM do I need for Flux? FLUX.1 [dev] loads 34.2 GB with everything in BF16/FP16, 29.3 GB with the T5 text encoder in FP8 and 17.2 GB as the all-in-one FP8 file. That means an L40S, an RTX 5090 or an RTX 4090. FLUX.2 [dev] loads 112.8 GB in BF16. ### Can I run FLUX.1 [dev] on a 24 GB RTX 4090? Yes. The FP8 checkpoint (17.2 GB) fits entirely. The BF16 model (23.8 GB) also runs, because ComfyUI moves weights between system RAM and VRAM, but each new prompt then waits for those transfers. ### How much VRAM does SDXL need? The SDXL 1.0 base checkpoint is 6.9 GB with its text encoders and VAE, so every GPU we rent holds it. An RTX 4090 at $0.28/hour leaves room for LoRAs, ControlNet models and upscalers in the same graph. ### How much VRAM does FLUX.2 [dev] need? Its 32B model takes 64.4 GB in BF16 and its text encoder 48.0 GB, 112.8 GB in all. One H200 holds both; an 80 GB H100 holds the model with the text encoder unloaded. The model card also shows a 4-bit version on an RTX 4090 with a remote text encoder. ### Can I use FLUX models commercially? Check each licence. FLUX.1 [schnell] is Apache-2.0. FLUX.1 [dev], FLUX.2 [dev] and FLUX.2 [klein] 9B use Black Forest Labs' non-commercial licences; read their terms before commercial work. ### Can I pay for the GPU with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR from a $1 top-up, with no ID and no card. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. ## Sources 1. [FLUX.1 [dev] model card and files](https://huggingface.co/black-forest-labs/FLUX.1-dev) — Black Forest Labs / Hugging Face (read 2026-10-29) 2. [FLUX.1 [schnell] model card and files](https://huggingface.co/black-forest-labs/FLUX.1-schnell) — Black Forest Labs / Hugging Face (read 2026-10-29) 3. [FLUX.2 [dev] model card and files](https://huggingface.co/black-forest-labs/FLUX.2-dev) — Black Forest Labs / Hugging Face (read 2026-10-29) 4. [FLUX.2 [klein] 9B model card and files](https://huggingface.co/black-forest-labs/FLUX.2-klein-9B) — Black Forest Labs / Hugging Face (read 2026-10-29) 5. [Flux text encoders (T5-XXL, CLIP-L) for ComfyUI](https://huggingface.co/comfyanonymous/flux_text_encoders) — ComfyUI / Hugging Face (read 2026-10-29) 6. [Comfy-Org Flux-1-Dev files (FP8 checkpoint)](https://huggingface.co/Comfy-Org/Flux-1-Dev) — Comfy-Org / Hugging Face (read 2026-10-29) 7. [Stable Diffusion XL 1.0 announcement](https://stability.ai/news-updates/stable-diffusion-sdxl-1-announcement) — Stability AI (read 2026-10-29) 8. [stable-diffusion-xl-base-1.0 files](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) — Stability AI / Hugging Face (read 2026-10-29) 9. [ComfyUI README and command-line options](https://github.com/comfyanonymous/ComfyUI) — ComfyUI (read 2026-10-29) 10. [NVIDIA RTX Blackwell architecture whitepaper (RTX 5090)](https://images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdf) — NVIDIA (read 2026-10-29) 11. [NVIDIA Ada GPU architecture whitepaper (RTX 4090)](https://images.nvidia.com/aem-dam/Solutions/geforce/ada/nvidia-ada-gpu-architecture.pdf) — NVIDIA (read 2026-10-29) 12. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 13. [Vast.ai public GPU price feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What is the best GPU to fine-tune Llama 70B (LoRA, QLoRA, full)? > Best GPU to fine-tune Llama 70B: one 80 GB A100 or H100 for QLoRA (52.5 GB), one B200 for LoRA (168 GB), 8× B200 for full. Memory math and crypto prices. - URL: https://gpuwithcrypto.com/best-gpu-for/fine-tuning-llama-70b/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) The method decides the GPU. Here is the memory each method needs for a 70B model, which GPU counts hold it, and what a run costs under clearly stated assumptions. **Quick answer — What is the best GPU to fine-tune Llama 70B?** The best GPU to fine-tune Llama 70B depends on the method. QLoRA needs about 52.5 GB, so one 80 GB A100 or H100. LoRA on a BF16 base needs about 168 GB, so one 180 GB B200 or two H200s. Full fine-tuning needs 1,120 to 1,260 GB before activations, so eight B200s. As of 29 October 2026, **GPUwithCrypto** rents an A100 80GB at $0.92/hour, paid in crypto. ## Which GPU for which method? | Attribute | Value | | --- | --- | | QLoRA (4-bit base) | ≈ 52.5 GB → 1× A100 80GB ($0.92/hour) or 1× H100 ($2.32/hour) | | LoRA (BF16 base) | 140 GB floor, ≈ 168 GB to plan → 1× B200 ($4.61/hour) or 2× H200 ($6.14/hour) | | Full fine-tune, 8-bit Adam | ≈ 840 GB → 8× H200 ($24.56/hour) | | Full fine-tune, AdamW | 1,120–1,260 GB → 8× B200 ($36.88/hour) | | Too small | One 48 GB L40S misses the 52.5 GB QLoRA estimate; the QLoRA authors did fit a 65B model on one 48 GB GPU | ## How much memory does each method need for a 70B model? - **QLoRA:** the base model is loaded in 4-bit and frozen; only small adapters train. The QLoRA paper fine-tuned a 65B model on one 48 GB GPU: 48 ÷ 65 = 0.74 GB per billion parameters, rounded to 0.75. For 70B: 70 × 0.75 = **52.5 GB**. - **LoRA:** the base stays in BF16 and frozen: 70 × 2 bytes = **140 GB**. Adapters and their optimizer add about 1%; with our 20% planning margin, 140 × 1.2 = **168 GB**. - **Full fine-tuning, mixed precision with AdamW:** EleutherAI counts 2 + 12 + 2 = 16 bytes per parameter (weights, optimizer with FP32 copy, gradients): 70 × 16 = **1,120 GB**. Hugging Face counts 6 + 8 + 4 = 18 bytes: **1,260 GB**. Activations come on top. - **Full fine-tuning with 8-bit Adam:** Hugging Face puts 8-bit optimizer states at 2 bytes instead of 8, so 6 + 2 + 4 = 12 bytes: 70 × 12 = **840 GB**. These are planning rules from the cited sources, not measurements. Sequence length and batch size change the activation memory; gradient checkpointing reduces it. The same rules for other model sizes: [how much VRAM do I need?](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") ## How many GPUs does each method need? ✓ = the configuration's combined VRAM holds the method's estimate. Multi-GPU rows assume the model is sharded across GPUs (FSDP or DeepSpeed ZeRO-3). Our prices per hour for the whole configuration, 29 October 2026. | Configuration | Total VRAM | QLoRA (52.5 GB) | LoRA (168 GB) | Full, 8-bit Adam (840 GB) | Full, AdamW (1,260 GB) | GPUwithCrypto per hour | | --- | --- | --- | --- | --- | --- | --- | | 1× L40S | 48 GB | No | No | No | No | $0.82/hour | | 1× A100 80GB | 80 GB | Yes | No | No | No | $0.92/hour | | 1× H100 | 80 GB | Yes | No | No | No | $2.32/hour | | 1× H200 | 141 GB | Yes | No | No | No | $3.07/hour | | 1× B200 | 180 GB | Yes | Yes | No | No | $4.61/hour | | 2× H200 | 282 GB | Yes | Yes | No | No | $6.14/hour | | 4× A100 80GB | 320 GB | Yes | Yes | No | No | $3.68/hour | | 8× H100 | 640 GB | Yes | Yes | No | No | $18.56/hour | | 8× H200 | 1,128 GB | Yes | Yes | Yes | No | $24.56/hour | | 8× B200 | 1,440 GB | Yes | Yes | Yes | Yes | $36.88/hour | *Prices checked on October 29, 2026.* 8× H200 holds the 16-byte estimate (1,120 GB) on paper but leaves 8 GB for activations, so it is marked ✗ for AdamW. 8× B200 leaves 180 GB after the 18-byte estimate. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## What does a Llama 70B fine-tuning run cost? Cost = GPUs × wall-clock hours × price per GPU-hour. **The hours below are assumptions chosen for illustration, not measurements:** run time depends on the number of training tokens, sequence length, batch size and software. Measure your own steps per hour on a short run before you size a long one. The assumed hours match our blocks of 10, 24 and 168 GPU-hours. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. Every block and GPU is listed on [GPU pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ### Run-cost estimate under assumed hours | Method | Configuration | Assumed hours (not measured) | GPU-hours | GPUwithCrypto cost | | --- | --- | --- | --- | --- | | QLoRA | 1× A100 80GB | 10 | 10 | $9.20 | | QLoRA | 1× H100 | 10 | 10 | $23.20 | | LoRA | 1× B200 | 24 | 24 | $110.64 | | LoRA | 2× H200 | 12 | 24 | $73.68 | | Full, 8-bit Adam | 8× H200 | 21 | 168 | $515.76 | | Full, AdamW | 8× B200 | 21 | 168 | $774.48 | *Prices checked on October 29, 2026.* ## How much more does full fine-tuning cost than QLoRA? About 40 times more per hour of hardware. According to GPUwithCrypto's price check of 29 October 2026, the lowest self-serve on-demand price for an 8× B200 instance was $43.46 per hour (Massed Compute, $5.4325 per GPU). One A100 80GB cost $1.09 per hour at Thunder Compute. The ratio is the same at our prices: $36.88/hour against $0.92/hour. That gap is why most 70B projects start with QLoRA or LoRA. The LoRA paper reports results on par with full fine-tuning on the tasks it tested, with 10,000 times fewer trainable parameters. ## Is the H100 or the A100 better for QLoRA on 70B? Both hold the 52.5 GB estimate in 80 GB. The H100 has 3.2 times the A100's peak BF16 Tensor throughput (989.5 against 312 TFLOPS dense) and 3,350 GB/s of bandwidth (SXM) against 2,039 GB/s. These are NVIDIA's peak figures; real training speed depends on the software. The A100 costs $0.92/hour and the H100 $2.32/hour at our prices. Pick the A100 when the budget is fixed, the H100 when the deadline is. More: [H100 vs A100](https://gpuwithcrypto.com/compare/h100-vs-a100/ "H100 vs A100: which should I rent?"). ## Which software runs each method? Hugging Face TRL runs QLoRA and LoRA from one command: `--load_in_4bit --use_peft` for QLoRA, `--use_peft` alone for LoRA. For multi-GPU LoRA and full fine-tuning, TRL ships ready-made Accelerate profiles such as `fsdp2` and `zero3` (DeepSpeed ZeRO-3), passed with `--accelerate_config`. The Llama 3.1 repositories on Hugging Face are gated: accept Meta's licence on the model page first. Step-by-step commands: [fine-tune Llama on a rented GPU](https://gpuwithcrypto.com/guides/fine-tune-llama-on-rented-gpu/ "How do I fine-tune Llama on a rented GPU?"). ## Llama 70B fine-tuning questions ### Can I fine-tune Llama 70B on one GPU? Yes, with QLoRA on an 80 GB GPU: the estimate is 52.5 GB. LoRA on a BF16 base also fits one GPU, but only a 180 GB B200. Full fine-tuning needs eight B200s or eight H200s with an 8-bit optimizer. ### Can I fine-tune Llama 70B on an RTX 4090? No. Even in 4-bit, the 70B weights alone take 35 GB, more than the RTX 4090's 24 GB. The smallest GPU we rent that holds a QLoRA run is an 80 GB A100. ### How much VRAM do I need to fine-tune Llama 70B with LoRA? About 140 GB for the frozen BF16 weights, and about 168 GB with a 20% margin for adapters and activations. One B200 (180 GB), two H200s (282 GB) or four 80 GB GPUs hold it. ### How long does fine-tuning Llama 70B take? It depends on your dataset's token count, sequence length and batch size, and we have not measured a run, so we publish no timing. Start with a 10-hour block, measure steps per hour on your data, then size the full run. ### Can I pay for the GPUs with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR from a $1 top-up, with no ID and no card. Unused GPU-hours never expire. ## Sources 1. [QLoRA: Efficient Finetuning of Quantized LLMs](https://arxiv.org/abs/2305.14314) — arXiv (read 2026-10-29) 2. [Making LLMs even more accessible with bitsandbytes, 4-bit quantization and QLoRA](https://huggingface.co/blog/4bit-transformers-bitsandbytes) — Hugging Face (read 2026-10-29) 3. [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) — arXiv (read 2026-10-29) 4. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) 5. [Model training anatomy (GPU memory)](https://huggingface.co/docs/transformers/model_memory_anatomy) — Hugging Face (read 2026-10-29) 6. [Llama 3.1 70B model card](https://huggingface.co/meta-llama/Llama-3.1-70B) — Meta / Hugging Face (read 2026-10-29) 7. [TRL command line interfaces](https://huggingface.co/docs/trl/clis) — Hugging Face (read 2026-10-29) 8. [TRL PEFT integration](https://huggingface.co/docs/trl/peft_integration) — Hugging Face (read 2026-10-29) 9. [NVIDIA A100 specifications](https://www.nvidia.com/en-us/data-center/a100/) — NVIDIA (read 2026-10-29) 10. [NVIDIA H100 specifications](https://www.nvidia.com/en-us/data-center/h100/) — NVIDIA (read 2026-10-29) 11. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 12. [Thunder Compute pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What is the best GPU for LLM inference? > The best GPU for LLM inference depends on model size: an RTX 4090 for 8B, an L40S for 32B in FP8, an H200 for 70B in FP8. Paid in crypto, no KYC, Oct 2026. - URL: https://gpuwithcrypto.com/best-gpu-for/llm-inference/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Pick the cheapest GPU whose memory holds the model plus its cache. The table below does that for every common model size, with our hourly price next to the lowest on-demand price we found elsewhere. **Quick answer — What is the best GPU for LLM inference?** The best GPU for LLM inference is the cheapest one whose VRAM holds the model plus about 20%: an RTX 4090 (24 GB) for 7–8B models, an L40S (48 GB) for 32B in FP8 or 70B in 4-bit, an H200 (141 GB) for 70B in FP8, and 8× H200 for 671B DeepSeek. As of 29 October 2026, **GPUwithCrypto** rents the H200 at $3.07/hour, paid in crypto. ## Which GPU should I use for each model size? VRAM needed = weights × 1.2 (see the [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?")). Our price and the lowest self-serve on-demand price per GPU-hour that we found among the providers we checked on 29 October 2026. | Model size (example) | VRAM needed | Best GPU | Why | GPUwithCrypto | Lowest elsewhere (per GPU) | | --- | --- | --- | --- | --- | --- | | 7–8B in BF16 (Llama 3.1 8B, Mistral 7B) | 17.5–19.2 GB | [RTX 4090 24GB](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") | Holds BF16 weights plus a 32K-token cache (4.3 GB on Llama 3.1 8B); FP8 Tensor Cores | $0.28/hour | $0.33 (Salad) | | 7–8B with a 128K-token context | 16 GB + 17.2 GB cache = 33.2 GB | [L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | One 128K-token BF16 cache alone needs 17.2 GB | $0.82/hour | $0.97 (Massed Compute) | | 30–32B in FP8 (Qwen2.5 32B) | 39 GB | [L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | 48 GB with FP8 Tensor Cores (733 TFLOPS dense FP8 peak) | $0.82/hour | $0.97 (Massed Compute) | | 30–32B in BF16 | 78 GB | [A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | Entry 80 GB GPU in our range; no FP8 Tensor Cores, so serve BF16 or 4-bit | $0.92/hour | $1.09 (Thunder Compute) | | 70–72B in 4-bit | 42–43.6 GB | [L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | Entry GPU above 44 GB in our range | $0.82/hour | $0.97 (Massed Compute) | | 70–72B in FP8 (Llama 3.1 70B, Qwen2.5 72B) | 84–87.2 GB | [H200 141GB](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") | One GPU instead of two H100s; 54–57 GB left for KV cache; 4,800 GB/s | $3.07/hour | $3.62 (Massed Compute) | | gpt-oss-120b (MXFP4) | 78.3 GB | [H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | Its model card targets a single 80 GB GPU; FP8 Tensor Cores, 3,350 GB/s (SXM) | $2.32/hour | $2.73 (Massed Compute) | | DeepSeek V4 Flash (284B MoE) | 191.5 GB | 2× H200 | Its FP4/FP8 weights take 159.6 GB; SGLang documents 4× H200 | $6.14/hour | $3.62 (Massed Compute) | | 405B in FP8 (Llama 3.1 405B) | 486 GB | 4× H200 | 564 GB on four GPUs; 8× H100 would need twice the GPU count | $12.28/hour | $3.62 (Massed Compute) | | 671B–1.6T MoE (DeepSeek-V3/R1 FP8, V4 Pro) | 805–1,037.6 GB | 8× H200 | 1,128 GB on one host; 8× H100 holds only 640 GB | $24.56/hour | $3.62 (Massed Compute) | *Prices checked on October 29, 2026.* Per GPU-hour, USD, taxes excluded. Spot and marketplace prices are not counted: they can be lower, but the GPU can be taken back. ## Why does VRAM matter more than TFLOPS for inference? VRAM decides whether the model runs at all, and memory bandwidth sets the speed limit for generating text. Each new token reads every active weight from memory once, so at small batch sizes the GPU waits on memory, not on arithmetic. Compute matters more for long prompts (prefill) and for large batches of users. That is why an H200 and an H100 have the same peak FP8 compute (1,979 TFLOPS dense, NVIDIA figures) but the H200 holds 76% more memory and moves it 43% faster: 4,800 GB/s against 3,350 GB/s. Bandwidth and TFLOPS here are NVIDIA's peak figures, not measurements of ours. ## Which GPU fits which model? (ranked by VRAM) **Methodology.** Published by GPUwithCrypto (SerpGoat LLC), which rents all seven GPUs. Each GPU is judged on the same three criteria. VRAM and bandwidth are NVIDIA's official figures; no speed was measured by us. Prices are ours and the lowest self-serve on-demand price we found on 29 October 2026. *Prices checked on October 29, 2026.* The seven GPUs we rent, listed by VRAM per GPU, largest first, because VRAM decides which model fits. Bandwidth is the second criterion, the price per hour the third. ### 1. NVIDIA B200 — Most VRAM 180 GB of HBM3e at 8,000 GB/s (per-GPU share of NVIDIA's 8-GPU HGX figures), FP8 and FP4 Tensor Cores. Holds a 70B model in BF16 (168 GB) on one GPU. - Best for: 70B in BF16 on one GPU, FP4 inference - Price: $4.61/hour · lowest elsewhere $5.43 ### 2. NVIDIA H200 141GB — Best for 70B 141 GB of HBM3e at 4,800 GB/s. Holds a 70B model in FP8 (84 GB) with 57 GB left for the KV cache. Eight of them hold DeepSeek-V3 or V4 Pro. - Best for: 70B in FP8, long contexts, large MoE models - Price: $3.07/hour · lowest elsewhere $3.62 ### 3. NVIDIA H100 80GB — All-round 80 GB of HBM3 at 3,350 GB/s (SXM) or 2,000 GB/s (PCIe), FP8 Tensor Cores. Holds gpt-oss-120b, or a 32B model in BF16. - Best for: gpt-oss-120b, 32B in BF16, high-throughput 8B serving - Price: $2.32/hour · lowest elsewhere $2.73 ### 4. NVIDIA A100 80GB — Entry 80 GB 80 GB of HBM2e at 2,039 GB/s (SXM) or 1,935 GB/s (PCIe). No FP8 Tensor Cores: serve BF16 or 4-bit. - Best for: 32B in BF16, 70B in 4-bit with long context - Price: $0.92/hour · lowest elsewhere $1.09 ### 5. NVIDIA L40S 48GB — Best value 48 GB 48 GB of GDDR6 at 864 GB/s, FP8 Tensor Cores, no NVLink. Holds 32B in FP8 or 70B in 4-bit. - Best for: 32B in FP8, 70B in 4-bit, 8B with 128K context - Price: $0.82/hour · lowest elsewhere $0.97 ### 6. NVIDIA GeForce RTX 5090 32 GB of GDDR7 at 1,792 GB/s, FP8 and FP4 Tensor Cores. Holds a 12B model in BF16 (28.8 GB) or a 32B model in 4-bit (19.5 GB) with room for a long context. - Best for: 12B in BF16, 32B in 4-bit with long prompts - Price: $0.42/hour · lowest elsewhere $0.50 ### 7. NVIDIA GeForce RTX 4090 — Entry GPU 24 GB of GDDR6X at 1,008 GB/s, FP8 Tensor Cores. Holds a 7–8B model in BF16. - Best for: 7–8B models, development and testing - Price: $0.28/hour · lowest elsewhere $0.33 *Disclosure: this page is published by GPUwithCrypto. Competitor details come from their public websites at the date shown above and may have changed since.* ## Is one H200 better value than two H100s? For a 70B model in FP8, yes: one H200 holds it, while two H100s must split it and exchange data on every layer. According to GPUwithCrypto's price check of 29 October 2026, the median on-demand H200 across 13 providers was $5.29 per GPU-hour, against $4.16 for an H100 across 16 providers. The H200 costs 27% more per hour for 76% more VRAM, so it is 28% cheaper per GB of memory. At our prices the choice is $3.07/hour for one H200 against $4.64/hour for two H100s. Full comparison: [H100 vs H200](https://gpuwithcrypto.com/compare/h100-vs-h200/ "H100 vs H200: is the H200 worth the extra cost?"). ## What about spot and marketplace GPUs? They cost less and can disappear mid-job. The cheapest H100 offer on Vast.ai's marketplace was $1.73 per hour on 29 October 2026 (Vast.ai public price feed), below the lowest on-demand price we found. For an inference server that must stay up, an interruption means downtime. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. When to use each: [what is a spot GPU](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?"). ## LLM inference GPU questions ### Which GPU do I need to run Llama 70B? One H200 in FP8 (84 GB needed, 141 GB available), one L40S in 4-bit (42 GB needed, 48 GB available) or one B200 in BF16 (168 GB needed, 180 GB available). Two H100s also hold the FP8 version, at 160 GB combined. ### What is the best GPU for DeepSeek V4? The H200. DeepSeek V4 Flash's weights take 159.6 GB, so it needs at least two H200s, and the SGLang cookbook documents four. V4 Pro's 864.7 GB need eight H200s. Commands are in the [DeepSeek V4 guide](https://gpuwithcrypto.com/guides/run-deepseek-v4-on-cloud-gpu/ "How do I run DeepSeek V4 on a cloud GPU?"). ### Is the RTX 4090 good for LLM inference? For 7–8B models, yes: 24 GB holds Llama 3.1 8B in BF16 (19.2 GB with overhead) and it is the entry GPU of our range, with 1,008 GB/s of bandwidth. Above about 10B parameters in BF16 it runs out of memory. ### Is the A100 still good for inference in 2026? For BF16 and 4-bit models up to 80 GB, yes, and it is the entry 80 GB GPU in our range. It has no FP8 Tensor Cores, so FP8 checkpoints run better on an L40S, H100 or H200. See [H100 vs A100](https://gpuwithcrypto.com/compare/h100-vs-a100/ "H100 vs A100: which should I rent?"). ### Can I rent these GPUs with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR from a $1 top-up, with no ID and no card. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. ## Sources 1. [NVIDIA H200 specifications](https://www.nvidia.com/en-us/data-center/h200/) — NVIDIA (read 2026-10-29) 2. [NVIDIA H100 specifications](https://www.nvidia.com/en-us/data-center/h100/) — NVIDIA (read 2026-10-29) 3. [NVIDIA A100 specifications](https://www.nvidia.com/en-us/data-center/a100/) — NVIDIA (read 2026-10-29) 4. [NVIDIA L40S specifications](https://www.nvidia.com/en-us/data-center/l40s/) — NVIDIA (read 2026-10-29) 5. [NVIDIA HGX platform (B200)](https://www.nvidia.com/en-us/data-center/hgx/) — NVIDIA (read 2026-10-29) 6. [NVIDIA Ada GPU architecture whitepaper (RTX 4090)](https://images.nvidia.com/aem-dam/Solutions/geforce/ada/nvidia-ada-gpu-architecture.pdf) — NVIDIA (read 2026-10-29) 7. [NVIDIA RTX Blackwell architecture whitepaper (RTX 5090)](https://images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdf) — NVIDIA (read 2026-10-29) 8. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) 9. [gpt-oss-120b model card](https://huggingface.co/openai/gpt-oss-120b) — OpenAI / Hugging Face (read 2026-10-29) 10. [DeepSeek-V4 cookbook](https://docs.sglang.io/cookbook/autoregressive/DeepSeek/DeepSeek-V4) — SGLang (read 2026-10-29) 11. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 12. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 13. [Thunder Compute pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 14. [Vast.ai public GPU price feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What is the cheapest GPU cloud for AI right now? > Cheapest GPU cloud for AI on 29 Oct 2026, per GPU: lowest on-demand prices (H100 $2.73, RTX 4090 $0.33), spot listed apart, plus crypto and KYC rules. - URL: https://gpuwithcrypto.com/cheapest-gpu-cloud/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) There is no single cheapest GPU cloud: the cheapest provider changes with the GPU, and spot or marketplace offers undercut every on-demand price. Here is the ranking per GPU, with the catches. **Quick answer — What is the cheapest GPU cloud for AI right now?** The cheapest GPU cloud for AI depends on the GPU. As of 29 October 2026, the lowest self-serve on-demand prices we found were Massed Compute for the H100 ($2.73), H200, B200 and L40S, Thunder Compute for the A100 80GB ($1.09), and Salad for the RTX 4090 ($0.33) and RTX 5090. **GPUwithCrypto** prices each GPU at least 15% under a named on-demand reference, paid in crypto: H100 $2.32/hour. ## Which cloud has the cheapest on-demand price for each GPU? According to GPUwithCrypto's price check of 29 October 2026: the lowest self-serve on-demand price per GPU-hour, the next provider, and the median of all providers listing that GPU. GPUwithCrypto prices each GPU at least 15% under a named on-demand reference read the same day. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. | GPU | Cheapest on-demand | Runner-up | Median (providers) | GPUwithCrypto | Crypto at the cheapest | | --- | --- | --- | --- | --- | --- | | [H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | $2.73 (Massed Compute) | $2.89 (RunPod, PCIe) | $4.16 (16) | **$2.32/hour** | Yes: BTC | | [H200 141GB](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") | $3.62 (Massed Compute) | $3.99 (Hyperbolic) | $5.29 (13) | **$3.07/hour** | Yes: BTC | | B200 | $5.43 (Massed Compute, 8-GPU node ÷ 8) | $5.99 (Hyperbolic) | $8.50 (9) | **$4.61/hour** | Yes: BTC | | [A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | $1.09 (Thunder Compute) | $1.35 (Massed Compute) | $2.23 (13) | **$0.92/hour** | No: card or preloaded credit | | [L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | $0.97 (Massed Compute) | $1.09 (RunPod) | $1.67 (11) | **$0.82/hour** | Yes: BTC | | [RTX 4090](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") | $0.33 (Salad) | $0.89 (RunPod) | $0.61 (2) | **$0.28/hour** | Yes: USDC, USDT, RENDER | | [RTX 5090](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") | $0.50 (Salad) | $0.84 (Spheron) | $0.84 (3) | **$0.42/hour** | Yes: USDC, USDT, RENDER | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded. Salad's High Priority prices rise on 1 November 2026: RTX 4090 to $0.44, RTX 5090 to $0.60. Voltage Park's "starting at $1.99" H100 is not counted: its pricing cards say "Contact for pricing". The full provider list is in the [GPU rental price index](https://gpuwithcrypto.com/gpu-price-index/ "GPU rental price index: what did a GPU-hour cost in October 2026?"). ## Is a spot or marketplace GPU cheaper? Yes, for every GPU. These offers can be taken back, change with each host, or run on community machines, so they are listed apart from on-demand prices. Each is labelled with what the provider calls it. | GPU | Lowest spot / marketplace | What it is | Other low offers | | --- | --- | --- | --- | | H100 80GB | $0.79 (Nebius) | Preemptible "from" price, dynamic | Jarvislabs spot $1.54; RunPod Community Cloud $1.99 (PCIe); Akash bids from $2.04 | | H200 141GB | $0.79 (Nebius) | Preemptible "from" price, dynamic | Jarvislabs spot $2.29; CoreWeave spot $2.62 | | B200 | $0.99 (Nebius) | Preemptible "from" price, dynamic | CoreWeave spot $4.26; Verda spot $4.39 | | A100 80GB | $0.89 (Jarvislabs) | Spot, interruptible | Akash bids from $1.07; Verda spot $1.11; RunPod Community Cloud $1.19 | | L40S | $0.74 (Nebius) | Preemptible "from" price, dynamic | RunPod Community Cloud $0.79; Verda spot $0.94 | | RTX 4090 | $0.16 (Salad) | Batch priority, "most interruptible" | RunPod Community Cloud $0.34; Akash bid $0.34 | | RTX 5090 | $0.25 (Salad) | Batch priority, "most interruptible" | Akash bid $0.57 (none available); RunPod Community Cloud $0.69 | *Prices checked on October 29, 2026.* Multi-GPU prices divided by the GPU count. Vast.ai is quoted separately below: its price-feed licence asks for a data licence before use in a benchmark. ## When is spot or marketplace cheaper than GPUwithCrypto? Always, on 29 October 2026: for all seven GPUs, at least one spot, preemptible or marketplace offer was below the GPUwithCrypto price. Vast.ai's public price feed went lower still, with an RTX 4090 from $0.13 per hour. What you trade is certainty. Nebius says its highest preemptible rate is always one cent below on-demand, so the "from" price moves. Salad's Batch tier is its "most interruptible". A marketplace host sets its own terms. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. Spot suits jobs that checkpoint often and restart cleanly. A long fine-tuning run or a live inference server is usually worth an on-demand GPU. The [spot GPU explainer](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?") covers the trade-off in detail. ## Which of the cheapest GPU clouds accept crypto? Most of them. Massed Compute, cheapest on-demand for the H100, H200, B200 and L40S, takes Bitcoin, with no ID step documented, but has no auto-recharge for crypto. Salad, cheapest for the RTX 4090 and RTX 5090, takes USDC, USDT and RENDER on Solana; crypto payments cannot be refunded. Thunder Compute, cheapest for the A100 80GB, lists only a saved card or preloaded credit in its billing docs. RunPod, runner-up for the H100, L40S and RTX 4090, takes crypto after KYC. Hyperbolic's crypto support could not be confirmed. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID, and prices each GPU at least 15% under a named on-demand reference. The [crypto comparison of 33 GPU clouds](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") has the full rules. ## What should I check besides the hourly price? Billing increments, minimums and data charges change the real cost of a short job. Facts from each provider's own pages, read on 29 October 2026; "—" means not stated. | Provider | Billing increment | Minimum | Egress / storage | Payment / ID | | --- | --- | --- | --- | --- | | **GPUwithCrypto** (ours) | Per minute | $1 top-up | Local NVMe included, wiped on delete | BTC, USDT, USDC, ETH, SOL, LTC, XMR; no ID | | Massed Compute | Hourly rate, increment not stated | No long-term commitment | "No Bandwidth Charges"; local storage included | Card or BTC; no ID step documented | | RunPod | Per second | $10 to start | Container disk $0.10/GB-month; idle volume $0.20/GB-month | Card, or crypto after KYC | | Thunder Compute | Per minute | — | No egress fees; 100 GB disk per GPU, then $0.219/GB-month | Card or preloaded credit | | Salad | Per second | $5 | — | Card or USDC, USDT, RENDER | | Jarvislabs | Per minute | $10 | Free in-region egress; disks $0.10/GB-month | Card; bank transfer above $500 | | Lambda | Per minute | None for instances | Local SSD included | Major credit cards only | | Crusoe | Per hour or per second | No minimum commitment | No ingress or egress charges; persistent disk $0.08/GiB-month | Non-prepaid credit card | | Nebius | — | $25 | Egress and ingress free | Card; bank transfer for companies | | Spheron | Per minute after a 20-minute minimum runtime | $15 by card, $50 by crypto | Bandwidth and NVMe included | Card or USDC | | Paperspace / DigitalOcean | Per second, 5-minute minimum | $25 authorization for crypto | 15,000 GiB transfer included per Droplet | Card, PayPal, stablecoins | | OVHcloud | Per second, 1-minute minimum | — | Local NVMe/SSD included | Card, PayPal; ID sometimes | | Google Cloud | Per second after a 1-minute minimum | Postpaid | Local SSD bundled | Card or bank; no crypto | *Prices checked on October 29, 2026.* ## Why are AWS, Google Cloud and Azure so expensive per GPU? Because their GPU price includes the whole VM, and most of their large GPUs come as 8-GPU machines. On 29 October 2026 an on-demand H100 cost $6.88 per GPU-hour at AWS (us-east-1), $11.06 at Google Cloud (Iowa) and $12.29 at Azure (East US), against $2.73 at Massed Compute. The [price index](https://gpuwithcrypto.com/gpu-price-index/ "GPU rental price index: what did a GPU-hour cost in October 2026?") shows the H100 median falling from $4.16 to $3.49 without them. ## Cheapest GPU cloud: questions ### What is the cheapest cloud GPU for AI? Per GPU-hour on-demand, the RTX 4090: $0.33 at Salad until 31 October 2026, then $0.44. GPUwithCrypto rents one for $0.28/hour. For 80 GB of VRAM, the cheapest on-demand was the A100 80GB at Thunder Compute, $1.09. ### What is the cheapest H100 cloud right now? On-demand, Massed Compute at $2.73 per GPU-hour on 29 October 2026, and GPUwithCrypto at $2.32/hour. Preemptible H100s started lower, from $0.79 at Nebius, but can be taken back. ### Is spot always cheaper than on-demand? On 29 October 2026, yes, for all seven GPUs we track. Spot and preemptible GPUs can be interrupted, so they suit jobs that save checkpoints. ### Which cheap GPU cloud takes crypto with no ID? Massed Compute (Bitcoin) and Salad (USDC, USDT, RENDER) describe no ID step and had the lowest on-demand prices for six of the seven GPUs. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID, at least 15% under a named on-demand reference for each GPU. ### Is the cheapest GPU cloud reliable? We did not measure uptime, so we do not rank on it. What we can compare is the type of offer: on-demand GPUs stay yours while you pay, while spot, preemptible and marketplace offers can end early. ## Sources 1. [Pricing (VM)](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Crypto: automatic payments](https://vm-docs.massedcompute.com/docs/billing/crypto/automatic-payments) — Massed Compute (read 2026-10-29) 3. [Pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 4. [Billing](https://www.thundercompute.com/docs/billing.md) — Thunder Compute (read 2026-10-29) 5. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 6. [Billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 7. [Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 8. [GPU marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29) 9. [Prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 10. [Pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 11. [Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 12. [Pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 13. [GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) 14. [Public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 15. [Pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 16. [Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29) 17. [Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 18. [GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 19. [Public Cloud prices](https://www.ovhcloud.com/en/public-cloud/prices/) — OVHcloud (read 2026-10-29) 20. [Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 21. [EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — AWS (read 2026-10-29) 22. [Azure Retail Prices API, East US](https://prices.azure.com/api/retail/prices?$filter=armRegionName%20eq%20'eastus'%20and%20serviceName%20eq%20'Virtual%20Machines') — Microsoft (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # H100 vs A100: which should I rent? > H100 vs A100: which should I rent? The H100 has 3.2x the dense BF16 peak; the A100 80GB costs 60% less per hour. Specs, model fit, job cost, crypto prices. - URL: https://gpuwithcrypto.com/compare/h100-vs-a100/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Both GPUs have 80 GB of memory, so they fit the same models. The H100 is faster on paper and runs FP8; the A100 80GB costs far less per hour. Here is how to choose, with NVIDIA's official figures and the prices of 29 October 2026. **Quick answer — H100 vs A100: which should I rent?** **H100 vs A100:** rent the A100 80GB for memory-bound inference and LoRA fine-tuning of models that fit in 80 GB; rent the H100 for compute-bound training, FP8, or jobs that must finish fast. As of 29 October 2026, **GPUwithCrypto** rents the H100 at $2.32/hour and the A100 80GB at $0.92/hour, paid in crypto with no KYC. The H100 pays off only if it runs your job about 2.5 times faster. ## How do the H100 and A100 specs compare? Official NVIDIA figures, read on 29 October 2026. Tensor figures are **dense** peak TFLOPS. NVIDIA prints H100 Tensor values only with sparsity, so the H100 dense values are NVIDIA's figure ÷ 2, marked (d). Peak figures are not benchmarks. | Spec | H100 SXM | H100 PCIe | A100 80GB SXM | A100 80GB PCIe | | --- | --- | --- | --- | --- | | Architecture | Hopper | Hopper | Ampere | Ampere | | Memory | 80 GB HBM3 | 80 GB HBM2e | 80 GB HBM2e | 80 GB HBM2e | | Memory bandwidth | 3,350 GB/s | 2,000 GB/s | 2,039 GB/s | 1,935 GB/s | | FP16 / BF16 Tensor, dense | 989.5 TFLOPS (d) | — (1) | 312 TFLOPS | 312 TFLOPS | | FP8 Tensor, dense | 1,979 TFLOPS (d) | — | N/A (2) | N/A (2) | | NVLink | 900 GB/s | 600 GB/s (bridge) | 600 GB/s | 600 GB/s (2-GPU bridge) | | PCIe | Gen5 | Gen5 | Gen4 | Gen4 | | Max power | 700 W | 350 W | 400 W | 300 W | | Launch | 2022 | 2022 | 2020 | 2021 | (d) NVIDIA prints 1,979 TFLOPS (BF16) and 3,958 TFLOPS (FP8) with sparsity; dense is half. The A100 sparse figure is 624 TFLOPS. (1) No current official page gives H100 PCIe Tensor values as dense or sparse; an NVIDIA blog cites 1,513 TFLOPS BF16 without saying which. (2) NVIDIA lists FP8 as not available on Ampere. ## Which is faster, the H100 or the A100? On paper, the H100 SXM does 3.17 times the dense BF16 math of an A100 (989.5 vs 312 TFLOPS) and has 1.64 times the memory bandwidth (3,350 vs 2,039 GB/s). These are NVIDIA's peak figures, not benchmarks. Use the two ratios as planning bounds, not as measured speed. - **Compute-bound work** (training, fine-tuning at large batch, long prompt prefill) tracks the TFLOPS ratio more closely. - **Memory-bound work** (token-by-token LLM decoding at small batch) tracks the bandwidth ratio more closely. - **FP8** runs on the H100's Tensor Cores at 1,979 TFLOPS dense. The A100 has no FP8 Tensor Cores, so an FP8 checkpoint gets no FP8 speed-up there. The H100 PCIe is a different card. Its 2,000 GB/s is close to the 1,935 GB/s of an A100 80GB PCIe. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## Which models fit on an H100 or an A100? The same models fit on both, because both have 80 GB. Rule of thumb: **memory ≈ parameters (billions) × bytes per parameter × 1.2**, with 2 bytes for BF16, 1 for FP8/INT8 and 0.5 for 4-bit. The 20% covers KV cache and activations at modest context and batch (EleutherAI). Long contexts need more: see the [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?"). | Model and precision | Memory needed | H100 80GB | A100 80GB | | --- | --- | --- | --- | | Llama 3.1 8B, BF16 inference | 19.2 GB | 1 GPU | 1 GPU | | Llama 3.1 8B, LoRA fine-tune (planning) | 19.2 GB | 1 GPU | 1 GPU | | Llama 3.1 8B, full fine-tune | 128–144 GB + activations | 2+ GPUs | 2+ GPUs | | Llama 3.1 70B, 4-bit inference | 42 GB | 1 GPU | 1 GPU | | Llama 3.1 70B, QLoRA fine-tune | 52.5 GB | 1 GPU | 1 GPU | | Llama 3.1 70B, FP8 inference | 84 GB | 2 GPUs | 2 GPUs, no FP8 speed-up | | Qwen2.5 72B, FP8 inference | 87.2 GB | 2 GPUs | 2 GPUs, no FP8 speed-up | | Llama 3.1 70B, BF16 inference | 168 GB | 3 GPUs | 3 GPUs | | Llama 3.1 405B, FP8 inference | 486 GB | 7 GPUs | 7 GPUs, no FP8 speed-up | Naive fit = the number of 80 GB cards whose total memory holds the estimate; it ignores tensor-parallel overhead and memory reserved by the driver. LoRA planning = 2 bytes per parameter × 1.2. QLoRA ≈ 0.75 GB per billion parameters, calibrated on the QLoRA paper's 65B model on one 48 GB GPU. Full fine-tune = 16–18 bytes per parameter (EleutherAI, Hugging Face). ## How much do the H100 and A100 cost per hour? Per GPU-hour, USD. 'Lowest' and 'median' are self-serve on-demand list prices from GPUwithCrypto's price check of 29 October 2026; spot and marketplace prices are excluded from both. Vast.ai is a marketplace, so its figure is shown apart. Every GPU price is on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). | GPU | GPUwithCrypto | Lowest on-demand elsewhere | Median on-demand | Providers checked | Vast.ai lowest offer | | --- | --- | --- | --- | --- | --- | | [H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | **$2.32/hour** | $2.73 (Massed Compute) | $4.16 | 16 | $1.73 (H100 SXM) | | [A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | **$0.92/hour** | $1.09 (Thunder Compute) | $2.23 | 13 | $0.48 (SXM), $0.34 (PCIe) | *Prices checked on October 29, 2026.* Massed Compute and Thunder Compute do not state the form factor of these cards. Vast.ai figures: Vast.ai's public price feed, cheapest rentable offer on 29 October 2026. Taxes excluded. ## What does a GB of VRAM cost per hour on each? According to GPUwithCrypto's price check of 29 October 2026, a GB of A100 memory costs 2.5 times less per hour than a GB of H100 memory. Computed as hourly price ÷ 80 GB: - **A100 80GB:** 1.36 cents per GB-hour at the lowest on-demand price ($1.09 ÷ 80), 2.79 cents at the median ($2.23 ÷ 80). - **H100 80GB:** 3.41 cents per GB-hour at the lowest price ($2.73 ÷ 80), 5.20 cents at the median ($4.16 ÷ 80). At the median the gap shrinks to 1.87 times. At GPUwithCrypto the same division gives $0.012 and $0.029, each GPU priced at least 15% under its named on-demand reference. Per dollar at the lowest prices, the H100 SXM gives 27% more peak BF16 math (362 vs 286 TFLOPS per $/hour). The A100 SXM gives 52% more bandwidth per dollar (1,871 vs 1,227 GB/s per $/hour). If your job needs memory more than math, the A100 is the cheaper 80 GB. ## What does one job cost on an H100 vs an A100? The H100 is cheaper per job only when it runs the job faster than its price premium. That break-even is 2.50 times at the lowest prices we found ($2.73 ÷ $1.09) and 1.87 times at the median ($4.16 ÷ $2.23). Both GPUwithCrypto prices sit at least 15% under their named references (Massed Compute at $2.73, Thunder Compute at $1.09), so our break-even is also about 2.5 times. **Worked example (assumption, not a benchmark):** a fine-tuning job takes 24 hours on one A100 80GB. On an H100 SXM it takes 7.6 hours if it gains the full 3.17× BF16 peak ratio, or 14.6 hours if it only gains the 1.64× bandwidth ratio. ### Cost of the same job (24 A100-hours assumed) | Price basis | A100 80GB, 24 h | H100, 7.6 h (3.17× faster) | H100, 14.6 h (1.64× faster) | | --- | --- | --- | --- | | GPUwithCrypto | $22.08 | $17.63 | $33.87 | | Lowest on-demand elsewhere | $26.16 | $20.75 | $39.86 | | Median on-demand | $53.52 | $31.62 | $60.74 | *Prices checked on October 29, 2026.* Hours are an assumption; speed ratios are NVIDIA SXM peak figures. GPUwithCrypto bills per minute, so a 7.6-hour run is billed 7.6 hours. ## When should I rent the H100 instead of the A100? Rent the H100 when at least one of these is true: - **You train or fine-tune at large batch.** Compute-bound work tracks the 3.17× peak ratio more closely, and that ratio is above the 2.5× break-even. - **You serve FP8 models.** Only the H100 has FP8 Tensor Cores (1,979 TFLOPS dense). - **You need several GPUs talking fast.** H100 SXM NVLink runs at 900 GB/s, against 600 GB/s on the A100. - **A deadline matters more than the bill.** Rent the A100 80GB when the model fits in 80 GB and the work is memory-bound: chat inference at small batch, LoRA and QLoRA runs, evaluations, notebooks. If 80 GB is too small, read [H100 vs H200](https://gpuwithcrypto.com/compare/h100-vs-h200/ "H100 vs H200: is the H200 worth the extra cost?"). For a cheaper 48 GB card, read [L40S vs A100](https://gpuwithcrypto.com/compare/l40s-vs-a100/ "L40S vs A100 for inference and fine-tuning: which should I rent?"). ## Questions about the H100 and the A100 ### Is the H100 worth it over the A100? Only when it runs your job more than about 2.5 times faster, the ratio of the lowest hourly prices on 29 October 2026. NVIDIA's peak figures give the H100 SXM 3.17 times the dense BF16 math and 1.64 times the bandwidth. Compute-bound training clears that bar at peak; bandwidth-bound inference does not. ### How much does an H100 cost per hour compared with an A100? On 29 October 2026 the lowest self-serve on-demand price we found was $2.73 for an H100 80GB and $1.09 for an A100 80GB; the medians were $4.16 and $2.23. GPUwithCrypto charges $2.32/hour and $0.92/hour, paid in crypto. ### Can an A100 run FP8 models? The memory math is the same, but NVIDIA lists FP8 as not available on Ampere, so the A100 has no FP8 Tensor Core speed-up. The H100 runs FP8 at 1,979 dense TFLOPS, half of NVIDIA's 3,958 sparse figure. ### Can I run Llama 3.1 70B on one H100 or A100? Only in 4-bit: about 42 GB with the × 1.2 rule, which fits one 80 GB card. In FP8 it needs about 84 GB (two cards) and in BF16 about 168 GB (three cards). One H200 with 141 GB holds the FP8 version. ### Can I rent an H100 or A100 with crypto and no KYC? Yes. GPUwithCrypto rents both by the minute, paid in BTC, USDT, USDC, ETH, SOL, LTC, XMR, with no ID, no selfie and no card. Of the 33 GPU providers we checked on 29 October 2026, 14 accept crypto, and RunPod asks for KYC before the first crypto payment. **Rent an H100 or an A100 with crypto** H100 $2.32/hour · A100 80GB $0.92/hour · billed per minute · no KYC. [Rent an H100](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") · [Rent an A100](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") ## Sources 1. [NVIDIA H100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/h100/) — NVIDIA (read 2026-10-29) 2. [NVIDIA H100 PCIe product brief (PB-11133-001)](https://www.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf) — NVIDIA (read 2026-10-29) 3. [NVIDIA announces Hopper architecture (HBM3)](https://nvidianews.nvidia.com/news/nvidia-announces-hopper-architecture-the-next-generation-of-accelerated-computing) — NVIDIA (read 2026-10-29) 4. [NVIDIA A100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/a100/) — NVIDIA (read 2026-10-29) 5. [NVIDIA A100 datasheet](https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-nvidia-us-2188504-web.pdf) — NVIDIA (read 2026-10-29) 6. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) 7. [Model memory anatomy](https://huggingface.co/docs/transformers/model_memory_anatomy) — Hugging Face (read 2026-10-29) 8. [QLoRA: Efficient Finetuning of Quantized LLMs](https://arxiv.org/abs/2305.14314) — arXiv (read 2026-10-29) 9. [Llama 3.1 70B model card](https://huggingface.co/meta-llama/Llama-3.1-70B) — Meta / Hugging Face (read 2026-10-29) 10. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 11. [Thunder Compute pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 12. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # H100 vs H200: is the H200 worth the extra cost? > H100 vs H200: the H200 has 141 GB vs 80 GB and 1.43x the bandwidth for 33% more per hour. When it saves money, with specs, model fit and crypto prices. - URL: https://gpuwithcrypto.com/compare/h100-vs-h200/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) The H200 is an H100 with more and faster memory: same Hopper architecture, same peak math on the SXM version. Whether it is worth the higher hourly price depends on one question: does your model fit in 80 GB? **Quick answer — H100 vs H200: is the H200 worth the extra cost?** **H100 vs H200:** the H200 is worth the extra cost when your model does not fit in 80 GB. It has 141 GB and 1.43 times the bandwidth, with the same SXM peak compute. As of 29 October 2026 it cost 33% more per hour at the lowest on-demand prices we found, yet Llama 3.1 70B in FP8 fits one H200 instead of two H100s. **GPUwithCrypto** rents both for crypto, no KYC: H200 $3.07/hour, H100 $2.32/hour. ## How do the H100 and H200 specs compare? Official NVIDIA figures, read on 29 October 2026. Tensor figures are **dense** peak TFLOPS. NVIDIA prints Hopper Tensor values only with sparsity, so dense = NVIDIA's figure ÷ 2, marked (d). Peak figures are not benchmarks. | Spec | H100 SXM | H100 PCIe | H200 SXM | H200 NVL (PCIe) | | --- | --- | --- | --- | --- | | Memory | 80 GB HBM3 | 80 GB HBM2e | 141 GB HBM3e | 141 GB HBM3e | | Memory bandwidth | 3,350 GB/s | 2,000 GB/s | 4,800 GB/s | 4,800 GB/s | | FP16 / BF16 Tensor, dense | 989.5 TFLOPS (d) | — | 989.5 TFLOPS (d) | 835.5 TFLOPS (d) | | FP8 Tensor, dense | 1,979 TFLOPS (d) | — | 1,979 TFLOPS (d) | 1,670.5 TFLOPS (d) | | NVLink | 900 GB/s | 600 GB/s (bridge) | 900 GB/s | 900 GB/s (2- or 4-way bridge) | | PCIe | Gen5 | Gen5 | Gen5 | Gen5 | | Max power | 700 W | 350 W | 700 W | 600 W | | Launch | 2022 | 2022 | 2023 | 2024 | (d) = NVIDIA's 'with sparsity' figure ÷ 2. NVIDIA labels the H200 table 'Preliminary specifications. May be subject to change.' No current official page gives H100 PCIe Tensor values as dense or sparse. ## Is the H200 faster than the H100? Not in raw math. The H200 SXM and the H100 SXM share the same dense peak: 989.5 TFLOPS BF16 and 1,979 TFLOPS FP8. The H200 changes the memory: 141 GB instead of 80 GB (1.76 times) and 4,800 GB/s instead of 3,350 GB/s (1.43 times). These are NVIDIA's peak figures, not benchmarks. - **Compute-bound jobs** (training, long prompt prefill) get no peak advantage from the H200. - **Memory-bound decoding** can gain up to the 1.43× bandwidth ratio at peak. - **Capacity is the big gain.** A model that needs two H100s runs on one H200, with no split across GPUs. The H200 NVL is a PCIe card with the same 141 GB, a 600 W limit and lower peaks (835.5 TFLOPS BF16 dense). Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## Which models fit on one H200 but not on one H100? Rule of thumb: **memory ≈ parameters (billions) × bytes per parameter × 1.2**, with 2 bytes for BF16, 1 for FP8 and 0.5 for 4-bit (EleutherAI). Long contexts add KV cache: Llama 3.1 70B uses 320 KiB per token in BF16, so one 131,072-token context adds 42.9 GB. Formula and more models: [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?"). | Model and precision | Memory needed | H100 80GB | H200 141GB | | --- | --- | --- | --- | | Llama 3.1 8B, BF16 | 19.2 GB | 1 GPU | 1 GPU | | Llama 3.1 70B, QLoRA fine-tune | 52.5 GB | 1 GPU | 1 GPU | | Llama 3.1 70B, FP8 | 84 GB | 2 GPUs | **1 GPU** | | Qwen2.5 72B, FP8 | 87.2 GB | 2 GPUs | **1 GPU** | | Llama 3.1 70B, FP8 + one 128K-token context | ≈ 113 GB | 2 GPUs | **1 GPU** | | Llama 3.1 70B, BF16 | 168 GB | 3 GPUs | 2 GPUs | | Llama 3.1 405B, 4-bit | 243 GB | 4 GPUs | 2 GPUs | | Llama 3.1 405B, FP8 | 486 GB | 7 GPUs | 4 GPUs | | DeepSeek-V3 / R1, FP8 (native) | ≈ 805 GB | 11 GPUs (over one 8-GPU node) | 6 GPUs (one 8-GPU node) | Naive fit = the number of cards whose total memory holds the estimate; it ignores tensor-parallel overhead and memory reserved by the driver. DeepSeek publishes V3 and R1 weights in FP8; all 671B parameters must be resident although 37B are active per token. KV cache figures use the Llama 3 paper's architecture (80 layers, 8 KV heads, head size 128). ## How much do the H100 and H200 cost per hour? Per GPU-hour, USD. 'Lowest' and 'median' are self-serve on-demand list prices from GPUwithCrypto's price check of 29 October 2026; spot and marketplace prices are excluded. Vast.ai is a marketplace, shown apart. All GPUs: [pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). | GPU | GPUwithCrypto | Lowest on-demand elsewhere | Median on-demand | Providers checked | Vast.ai lowest offer | | --- | --- | --- | --- | --- | --- | | [H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | **$2.32/hour** | $2.73 (Massed Compute) | $4.16 | 16 | $1.73 (H100 SXM) | | [H200 141GB](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") | **$3.07/hour** | $3.62 (Massed Compute, H200 NVL) | $5.29 | 13 | $3.00 (H200 NVL) | *Prices checked on October 29, 2026.* Massed Compute does not state the form factor of its $2.73 H100. Vast.ai figures: Vast.ai's public price feed, cheapest rentable offer on 29 October 2026. Taxes excluded. ## Is the H200 cheaper per GB of memory? Yes. According to GPUwithCrypto's price check of 29 October 2026, the H200 costs 33% more per hour than the H100 but 25% less per GB of memory. Computed as hourly price ÷ GB: - **H200:** 2.57 cents per GB-hour at the lowest on-demand price ($3.62 ÷ 141), 3.75 cents at the median ($5.29 ÷ 141). - **H100:** 3.41 cents per GB-hour at the lowest price ($2.73 ÷ 80), 5.20 cents at the median ($4.16 ÷ 80). At the median the H200 is 28% cheaper per GB. At GPUwithCrypto the same division gives $0.022 and $0.029, each GPU priced at least 15% under its named on-demand reference. For bandwidth, the lowest-priced H200 gives 8% more GB/s per dollar than the lowest-priced H100 (1,326 vs 1,227 GB/s per $/hour, with the H100 at its SXM figure). ## What does it cost to serve Llama 70B on H100s vs an H200? One H200 costs less than two H100s. **Worked example (assumption):** you serve Llama 3.1 70B in FP8 (≈ 84 GB) around the clock. That takes two H100s or one H200. At the lowest prices we found, one H200 costs 34% less than two H100s; at the median, 36% less. Two H100s do give twice the peak math of one H200. That only matters if your serving is compute-bound at large batch. ### Llama 3.1 70B in FP8: two H100s or one H200 | Price basis | 2 × H100, 24 h | 1 × H200, 24 h | 2 × H100, 720 h | 1 × H200, 720 h | | --- | --- | --- | --- | --- | | GPUwithCrypto | 2 × $55.68 | $73.68 | 2 × $1,670.40 | $2,210.40 | | Lowest on-demand elsewhere | $131.04 | $86.88 | $3,931.20 | $2,606.40 | | Median on-demand | $199.68 | $126.96 | $5,990.40 | $3,808.80 | *Prices checked on October 29, 2026.* Running hours are an assumption. GPU time only; storage and network extra where the provider charges them. ## When should I rent the H100 instead? When the model fits in 80 GB. Then the H200 must run your job 1.33 times faster to pay off at the lowest prices ($3.62 ÷ $2.73), or 1.27 times at the median ($5.29 ÷ $4.16). Our two prices keep about the same 1.33 ratio, because each sits at least 15% under its named reference ($3.62 and $2.73). - **Compute-bound work that fits in 80 GB:** the H100 wins. Peak math is equal, the price is not. - **Bandwidth-bound decoding that fits in 80 GB:** the H200's 1.43× peak bandwidth edge sits just above the 1.33× price gap. Test your own workload before you commit to a month. - **Fine-tuning 8B models, QLoRA on 70B, notebooks:** the H100, or a cheaper [A100 80GB](https://gpuwithcrypto.com/compare/h100-vs-a100/ "H100 vs A100: which should I rent?"). Rent the H200 for 70B-class FP8 serving on one GPU, long contexts, 405B models on fewer GPUs, and DeepSeek-V3 or R1 on a single 8-GPU node. ## Questions about the H100 and the H200 ### Is the H200 faster than the H100? Not for math: the SXM versions share the same dense peak, 989.5 TFLOPS BF16 and 1,979 TFLOPS FP8 (derived from NVIDIA's sparse figures). The H200 has 1.43 times the memory bandwidth and 1.76 times the memory, which helps memory-bound decoding and large models. ### How much more does an H200 cost per hour than an H100? On 29 October 2026 the lowest self-serve on-demand prices we found were $3.62 for an H200 and $2.73 for an H100, 33% apart. The medians were $5.29 and $4.16. GPUwithCrypto charges $3.07/hour and $2.32/hour. ### Can I run Llama 3.1 70B on a single H200? Yes in FP8: about 84 GB with the × 1.2 rule, leaving room for KV cache in 141 GB. In BF16 it needs about 168 GB, so two H200s. On H100s the FP8 version needs two cards. ### Should I rent two H100s or one H200? For a model between 80 and 141 GB, one H200: it costs 34% less than two H100s at the lowest prices of 29 October 2026 and avoids splitting the model. Two H100s make sense when you need twice the peak math. ### Can I pay for an H200 with crypto? Yes. GPUwithCrypto rents the H200 for BTC, USDT, USDC, ETH, SOL, LTC, XMR, with no KYC. Of the 13 providers with an on-demand H200 price on 29 October 2026, 4 accept crypto: Massed Compute, Paperspace/DigitalOcean, Spheron and RunPod, which asks for KYC first. **Rent an H200 or an H100 with crypto** H200 $3.07/hour · H100 $2.32/hour · billed per minute · no KYC. [Rent an H200](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") · [Rent an H100](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") ## Sources 1. [NVIDIA H200 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/h200/) — NVIDIA (read 2026-10-29) 2. [NVIDIA H200 NVL](https://blogs.nvidia.com/blog/hopper-h200-nvl/) — NVIDIA (read 2026-10-29) 3. [NVIDIA H100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/h100/) — NVIDIA (read 2026-10-29) 4. [NVIDIA H100 PCIe product brief (PB-11133-001)](https://www.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf) — NVIDIA (read 2026-10-29) 5. [The Llama 3 Herd of Models (Table 3)](https://arxiv.org/abs/2407.21783) — arXiv (read 2026-10-29) 6. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) 7. [DeepSeek-V3 model card](https://huggingface.co/deepseek-ai/DeepSeek-V3) — DeepSeek / Hugging Face (read 2026-10-29) 8. [Qwen2.5-72B-Instruct model card](https://huggingface.co/Qwen/Qwen2.5-72B-Instruct) — Qwen / Hugging Face (read 2026-10-29) 9. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 10. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # L40S vs A100 for inference and fine-tuning: which should I rent? > L40S vs A100 for inference and fine-tuning: 48 GB with FP8 vs 80 GB with 2.4x the bandwidth and NVLink. Specs, model fit, cost per job and crypto prices. - URL: https://gpuwithcrypto.com/compare/l40s-vs-a100/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) The L40S is a newer Ada Lovelace card with FP8 and 48 GB of GDDR6. The A100 80GB is an older Ampere card with 80 GB of HBM2e, far more bandwidth and NVLink. Their lowest on-demand prices were only 12% apart on 29 October 2026, so the right pick depends on the job. **Quick answer — L40S vs A100 for inference and fine-tuning?** **L40S vs A100 for inference and fine-tuning:** pick the L40S for FP8 inference and image or video generation on models that fit in 48 GB. Pick the A100 80GB for fine-tuning, models between 48 and 80 GB, and multi-GPU jobs: it has NVLink and up to 2.4 times the bandwidth. As of 29 October 2026, **GPUwithCrypto** rents the L40S at $0.82/hour and the A100 80GB at $0.92/hour, paid in crypto with no KYC. ## How do the L40S and A100 specs compare? Official NVIDIA figures, read on 29 October 2026. NVIDIA prints both dense and sparse Tensor values for these cards; the table shows **dense** peak TFLOPS. Peak figures are not benchmarks. | Spec | L40S | A100 80GB SXM | A100 80GB PCIe | | --- | --- | --- | --- | | Architecture | Ada Lovelace | Ampere | Ampere | | Memory | 48 GB GDDR6 | 80 GB HBM2e | 80 GB HBM2e | | Memory bandwidth | 864 GB/s | 2,039 GB/s | 1,935 GB/s | | FP16 / BF16 Tensor, dense | 362.05 TFLOPS | 312 TFLOPS | 312 TFLOPS | | FP8 Tensor, dense | 733 TFLOPS | N/A | N/A | | NVLink | No | 600 GB/s | 600 GB/s (2-GPU bridge) | | PCIe | Gen4 x16 | Gen4 | Gen4 | | Max power | 350 W | 400 W | 300 W | | Launch | 2023 | 2020 | 2021 | Sparse figures: L40S 733 TFLOPS BF16 and 1,466 TFLOPS FP8; A100 624 TFLOPS BF16. NVIDIA lists FP8 as not available on Ampere. The L40S page states 'NVLink Support: No'. ## Is the L40S faster than the A100? For math, slightly: the L40S has 362.05 dense BF16 TFLOPS against 312 (1.16 times), plus FP8 at 733 TFLOPS, which the A100 lacks. For memory, no: the A100 80GB SXM moves 2,039 GB/s, 2.36 times the L40S's 864 GB/s (2.24 times for the A100 PCIe). These are NVIDIA's peak figures, not benchmarks. - **Token-by-token LLM decoding** depends mostly on bandwidth, where the A100 leads. - **Image and video diffusion** and large-batch prefill are compute-heavy, which favors the L40S's higher peak and FP8. - **Multi-GPU work** favors the A100. The L40S has no NVLink, so its GPUs talk over PCIe Gen4. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## Which models fit in 48 GB, and which need 80 GB? Rule of thumb for inference: **memory ≈ parameters (billions) × bytes per parameter × 1.2**, with 2 bytes for BF16, 1 for FP8 and 0.5 for 4-bit (EleutherAI). LoRA planning = 2 bytes per parameter × 1.2. QLoRA ≈ 0.75 GB per billion parameters. Formula and more models: [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?"). | Model and task | Memory needed | L40S 48GB | A100 80GB | | --- | --- | --- | --- | | Mistral 7B, BF16 inference | 17.5 GB | 1 GPU | 1 GPU | | Llama 3.1 8B, BF16 inference | 19.2 GB | 1 GPU | 1 GPU | | Llama 3.1 8B, LoRA fine-tune | 19.2 GB | 1 GPU | 1 GPU | | FLUX.1 [dev], BF16 weights | 24 GB + text encoders | 1 GPU | 1 GPU | | FLUX.1 [dev], LoRA fine-tune | 28.8 GB | 1 GPU | 1 GPU | | Llama 3.1 70B, 4-bit inference | 42 GB | 1 GPU | 1 GPU | | Qwen2.5 72B, 4-bit inference | 43.6 GB | 1 GPU | 1 GPU | | Llama 3.1 70B, QLoRA fine-tune | 52.5 GB | 2 GPUs (borderline on 1) | **1 GPU** | | Llama 3.1 70B, FP8 inference | 84 GB | 2 GPUs, no NVLink | 2 GPUs, no FP8 speed-up | | Llama 3.1 70B, LoRA fine-tune | 168 GB | 4 GPUs | 3 GPUs | Naive fit = the number of cards whose total memory holds the estimate; it ignores multi-GPU overhead and memory reserved by the driver. QLoRA's authors report a 65B model fine-tuned on one 48 GB GPU, hence 'borderline'. FLUX.1 also loads T5-XXL, CLIP-L and a VAE, which can be offloaded to CPU between steps. ## How much do the L40S and A100 cost per hour? Per GPU-hour, USD. 'Lowest' and 'median' are self-serve on-demand list prices from GPUwithCrypto's price check of 29 October 2026; spot and marketplace prices are excluded. Vast.ai is a marketplace, shown apart. All GPUs: [pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). | GPU | GPUwithCrypto | Lowest on-demand elsewhere | Median on-demand | Providers checked | Vast.ai lowest offer | | --- | --- | --- | --- | --- | --- | | [L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | **$0.82/hour** | $0.97 (Massed Compute) | $1.67 | 11 | $0.33 | | [A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | **$0.92/hour** | $1.09 (Thunder Compute) | $2.23 | 13 | $0.48 (SXM), $0.34 (PCIe) | *Prices checked on October 29, 2026.* At the lowest prices the A100 80GB costs 12% more per hour than the L40S; at the median, about a third more. Vast.ai figures: Vast.ai's public price feed, cheapest rentable offer on 29 October 2026. Taxes excluded. ## Which is cheaper per GB of VRAM and per TFLOPS? The A100 is cheaper per GB, the L40S per TFLOPS. According to GPUwithCrypto's price check of 29 October 2026, computed as hourly price ÷ GB of memory: - **A100 80GB:** 1.36 cents per GB-hour at the lowest on-demand price ($1.09 ÷ 80), 2.79 cents at the median ($2.23 ÷ 80). - **L40S 48GB:** 2.02 cents per GB-hour at the lowest price ($0.97 ÷ 48), 3.48 cents at the median ($1.67 ÷ 48). A GB of A100 memory is 33% cheaper per hour at the lowest prices and 20% cheaper at the median. At GPUwithCrypto the same division is $0.012 and $0.017. Per dollar at the lowest prices, the L40S gives 30% more peak BF16 math (373 vs 286 TFLOPS per $/hour) and 756 FP8 TFLOPS per $/hour. The A100 SXM gives 2.1 times the bandwidth per dollar (1,871 vs 891 GB/s per $/hour). ## What does one job cost on an L40S vs an A100? **Worked examples (assumptions, not benchmarks):** 1. **Serving Llama 3.1 8B for a month (720 hours).** It fits one card of either type. The L40S is the cheaper choice if its throughput is enough for your traffic. 2. **A QLoRA fine-tune of Llama 3.1 70B planned at 24 hours.** It needs about 52.5 GB: one A100, or two L40S. Assuming the same 24 hours on both setups, the A100 costs 44% less at the lowest prices and 33% less at the median. ### Cost of the two jobs | Price basis | 8B serving, 720 h: L40S | 8B serving, 720 h: A100 | 70B QLoRA, 24 h: 2 × L40S | 70B QLoRA, 24 h: 1 × A100 | | --- | --- | --- | --- | --- | | GPUwithCrypto | $590.40 | $662.40 | 2 × $19.68 | $22.08 | | Lowest on-demand elsewhere | $698.40 | $784.80 | $46.56 | $26.16 | | Median on-demand | $1,202.40 | $1,605.60 | $80.16 | $53.52 | *Prices checked on October 29, 2026.* Hours are assumptions. GPU time only. ## Which should I rent for inference, and which for fine-tuning? **For inference**, rent the L40S when the model fits in 48 GB with room for its KV cache: 7B–8B models in BF16, 70B models in 4-bit, and image or video generation with FLUX or SDXL. FP8 checkpoints run on its FP8 Tensor Cores. Rent the A100 80GB when the model needs 48–80 GB, when long contexts or large batches fill the cache, or when decoding speed is limited by bandwidth. **For fine-tuning**, the A100 80GB is the safer default. QLoRA on a 70B model fits one card, and NVLink at 600 GB/s helps multi-GPU runs. LoRA on 7B–8B models and FLUX LoRAs fit either card. More: [best GPU for LLM inference](https://gpuwithcrypto.com/best-gpu-for/llm-inference/ "What is the best GPU for LLM inference?"), [best GPU for ComfyUI and Flux](https://gpuwithcrypto.com/best-gpu-for/comfyui-flux/ "What is the best cloud GPU for ComfyUI, Stable Diffusion and Flux?"), [best GPU to fine-tune Llama 70B](https://gpuwithcrypto.com/best-gpu-for/fine-tuning-llama-70b/ "What is the best GPU to fine-tune Llama 70B (LoRA, QLoRA, full)?"). ## Questions about the L40S and the A100 ### Is the L40S better than the A100 for inference? For models that fit in 48 GB and for FP8 or image generation, yes: it has 1.16 times the dense BF16 peak and FP8 Tensor Cores. For larger models or bandwidth-bound decoding, the A100 80GB leads with 80 GB and 2.36 times the bandwidth (SXM). These are NVIDIA peak figures, not benchmarks. ### Is the L40S good for fine-tuning? For LoRA on 7B–8B models and FLUX LoRAs, yes: they fit in 48 GB. QLoRA on a 70B model needs about 52.5 GB by our rule, borderline on one L40S. Multi-GPU fine-tuning runs over PCIe, because the L40S has no NVLink. ### How much does an L40S cost per hour compared with an A100? On 29 October 2026 the lowest self-serve on-demand prices we found were $0.97 for an L40S and $1.09 for an A100 80GB; the medians were $1.67 and $2.23. GPUwithCrypto charges $0.82/hour and $0.92/hour. ### Does the L40S support NVLink? No. NVIDIA's L40S page states 'NVLink Support: No'; GPUs talk over PCIe Gen4 x16 at 64 GB/s bidirectional. The A100 80GB has NVLink at 600 GB/s. ### Can I rent an L40S or an A100 with crypto? Yes. GPUwithCrypto rents both for BTC, USDT, USDC, ETH, SOL, LTC, XMR, with no ID and no card. Of the 11 providers with an on-demand L40S price on 29 October 2026, 5 accept crypto: Massed Compute, RunPod (KYC first), Paperspace/DigitalOcean, Vultr and Spheron. **Rent an L40S or an A100 with crypto** L40S $0.82/hour · A100 80GB $0.92/hour · billed per minute · no KYC. [Rent an L40S](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") · [Rent an A100](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") ## Sources 1. [NVIDIA L40S GPU](https://www.nvidia.com/en-us/data-center/l40s/) — NVIDIA (read 2026-10-29) 2. [NVIDIA A100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/a100/) — NVIDIA (read 2026-10-29) 3. [NVIDIA A100 datasheet](https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-nvidia-us-2188504-web.pdf) — NVIDIA (read 2026-10-29) 4. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) 5. [QLoRA: Efficient Finetuning of Quantized LLMs](https://arxiv.org/abs/2305.14314) — arXiv (read 2026-10-29) 6. [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) — arXiv (read 2026-10-29) 7. [FLUX.1 [dev] model card](https://huggingface.co/black-forest-labs/FLUX.1-dev) — Black Forest Labs / Hugging Face (read 2026-10-29) 8. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 9. [Thunder Compute pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 10. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # RTX 5090 vs H100 for AI: when is the cheap card enough? > RTX 5090 vs H100 for AI: 32 GB vs 80 GB, 4.7x less dense compute, about 5x cheaper per hour. When the cheap card is enough, with model fit and crypto prices. - URL: https://gpuwithcrypto.com/compare/rtx-5090-vs-h100/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) The GeForce RTX 5090 is a 32 GB consumer card; the H100 is an 80 GB data-center GPU. The H100 has far more peak math, but it costs about five times more per hour. For models that fit in 32 GB, the cheap card often wins on cost per job. **Quick answer — RTX 5090 vs H100 for AI: when is the cheap card enough?** **RTX 5090 vs H100 for AI:** the cheap card is enough when your model, context and batch fit in 32 GB: 7B–8B LLM inference and LoRA, SDXL and FLUX image generation. Rent the H100 for 70B models, NVLink training or anything over 32 GB. As of 29 October 2026 the H100 cost 5.5 times more per hour at the lowest on-demand prices we found. **GPUwithCrypto** rents the RTX 5090 at $0.42/hour, paid in crypto, no KYC. ## How do the RTX 5090 and H100 specs compare? Official NVIDIA figures, read on 29 October 2026. Tensor figures are **dense** peak TFLOPS. GeForce cards have two Tensor rates; the table uses FP32 accumulate, the rate mixed-precision training and FP8 inference kernels use. Peak figures are not benchmarks. | Spec | GeForce RTX 5090 | H100 SXM | H100 PCIe | | --- | --- | --- | --- | | Architecture | Blackwell | Hopper | Hopper | | Memory | 32 GB GDDR7 | 80 GB HBM3 | 80 GB HBM2e | | Memory bandwidth | 1,792 GB/s | 3,350 GB/s | 2,000 GB/s | | FP16 / BF16 Tensor, dense | 209.5 TFLOPS | 989.5 TFLOPS (d) | — | | FP8 Tensor, dense | 419 TFLOPS | 1,979 TFLOPS (d) | — | | FP4 Tensor, dense | 1,676 TFLOPS | — | — | | NVLink | Not listed by NVIDIA | 900 GB/s | 600 GB/s (bridge) | | PCIe | Gen5 | Gen5 | Gen5 | | Max power | 575 W (total graphics power) | 700 W | 350 W | | Launch | 2025 | 2022 | 2022 | (d) = NVIDIA prints H100 Tensor values with sparsity only; dense = ÷ 2. With FP16 accumulate the RTX 5090 doubles (FP16 419, FP8 838 TFLOPS dense). NVIDIA's '3,352 AI TOPS' for the 5090 is sparse FP4. No current official page gives H100 PCIe Tensor values as dense or sparse. ## How much faster is the H100 than the RTX 5090? At peak, the H100 SXM does 4.72 times the dense BF16 and FP8 math of an RTX 5090 (989.5 vs 209.5 and 1,979 vs 419 TFLOPS). It has 1.87 times the memory bandwidth (3,350 vs 1,792 GB/s) and 2.5 times the memory. These are NVIDIA's peak figures, not benchmarks. The RTX 5090 has one edge: FP4 Tensor Cores (1,676 TFLOPS dense), which Hopper lacks. They help only with models quantized to FP4. The price gap is bigger than the math gap. At the lowest on-demand prices we found, the H100 cost 5.46 times more per hour ($2.73 ÷ $0.50); at the median, 4.95 times ($4.16 ÷ $0.84). Per dollar, the RTX 5090 gives 16% more peak BF16 math at the lowest prices (419 vs 362 TFLOPS per $/hour) and 5% more at the median. ## Which models fit in 32 GB, and which need an H100? Rule of thumb for inference: **memory ≈ parameters (billions) × bytes per parameter × 1.2**, with 2 bytes for BF16, 1 for FP8 and 0.5 for 4-bit (EleutherAI). Long contexts add KV cache: 128 KiB per token for Llama 3.1 8B in BF16. Formula: [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?"). | Model and task | Memory needed | RTX 5090 32GB | H100 80GB | | --- | --- | --- | --- | | SDXL 1.0 base, FP16 | 7 GB weights | 1 GPU | 1 GPU | | FLUX.1 [dev], FP8 | 12 GB weights | 1 GPU | 1 GPU | | FLUX.1 [dev], BF16 | 24 GB weights + text encoders | 1 GPU, encoders offloaded to CPU | 1 GPU | | FLUX.1 [dev], LoRA fine-tune | 28.8 GB | 1 GPU, tight | 1 GPU | | Mistral 7B, BF16 | 17.5 GB | 1 GPU | 1 GPU | | Llama 3.1 8B, BF16 | 19.2 GB | 1 GPU | 1 GPU | | Llama 3.1 8B, LoRA fine-tune | 19.2 GB | 1 GPU | 1 GPU | | Llama 3.1 8B, BF16 + one 128K-token context | 16 + 17.2 = 33.2 GB | Only with an FP8 KV cache (≈ 24.6 GB) | 1 GPU | | Llama 3.1 70B, 4-bit | 42 GB | 2 GPUs | 1 GPU | | Llama 3.1 70B, QLoRA fine-tune | 52.5 GB | 2 GPUs | 1 GPU | | Llama 3.1 70B, FP8 | 84 GB | 3 GPUs | 2 GPUs | Naive fit = the number of cards whose total memory holds the estimate; it ignores multi-GPU overhead and memory reserved by the driver. FLUX.1 pipelines also load T5-XXL, CLIP-L and a VAE (not in the 12B count); they can be offloaded to CPU between steps. An FP8 KV cache halves the cache size. LoRA planning = 2 bytes per parameter × 1.2; QLoRA ≈ 0.75 GB per billion parameters. ## How much do the RTX 5090 and H100 cost per hour? Per GPU-hour, USD. 'Lowest' and 'median' are self-serve on-demand list prices from GPUwithCrypto's price check of 29 October 2026; spot and marketplace prices are excluded. Vast.ai is a marketplace, shown apart. All GPUs: [pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). | GPU | GPUwithCrypto | Lowest on-demand elsewhere | Median on-demand | Providers checked | Vast.ai lowest offer | | --- | --- | --- | --- | --- | --- | | [RTX 5090 32GB](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") | **$0.42/hour** | $0.50 (Salad, High Priority) | $0.84 | 3 | $0.39 | | [H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | **$2.32/hour** | $2.73 (Massed Compute) | $4.16 | 16 | $1.73 (H100 SXM) | *Prices checked on October 29, 2026.* Salad raises its RTX 5090 High Priority price to $0.60 on 1 November 2026, per its own price-change notice. The three RTX 5090 providers are Salad, Spheron ($0.84) and RunPod ($1.19). Vast.ai figures: Vast.ai's public price feed, cheapest rentable offer on 29 October 2026. ## What does a GB of VRAM cost per hour on each? According to GPUwithCrypto's price check of 29 October 2026, a GB of RTX 5090 memory costs 54% less per hour than a GB of H100 memory. Computed as hourly price ÷ GB: - **RTX 5090:** 1.56 cents per GB-hour at the lowest on-demand price ($0.50 ÷ 32), 2.63 cents at the median ($0.84 ÷ 32). - **H100:** 3.41 cents per GB-hour at the lowest price ($2.73 ÷ 80), 5.20 cents at the median ($4.16 ÷ 80). At the median the RTX 5090 costs about half as much per GB. At GPUwithCrypto the same division gives $0.013 and $0.029, each GPU priced at least 15% under its named on-demand reference. The catch: you cannot buy a GB beyond 32 on one 5090. ## What does one job cost on an RTX 5090 vs an H100? For a job that fits in 32 GB, the RTX 5090 costs less even if the H100 reaches its full peak advantage. The H100 is 4.72 times faster at peak but costs 5.46 times more per hour at the lowest prices and 4.95 times more at the median. Our two prices each sit at least 15% under their named references ($0.50 and $2.73), so the ratio at GPUwithCrypto is about the same. **Worked example (assumption, not a benchmark):** a FLUX LoRA run or an image batch takes 10 hours on one RTX 5090. On an H100 SXM it takes 2.1 hours at the full 4.72× math ratio, or 5.3 hours at the 1.87× bandwidth ratio. ### Cost of the same job (10 RTX 5090-hours assumed) | Price basis | RTX 5090, 10 h | H100, 2.1 h (4.72× faster) | H100, 5.3 h (1.87× faster) | | --- | --- | --- | --- | | GPUwithCrypto | $4.20 | $4.87 | $12.30 | | Lowest on-demand elsewhere | $5.00 | $5.73 | $14.47 | | Median on-demand | $8.40 | $8.74 | $22.05 | *Prices checked on October 29, 2026.* Hours are an assumption; speed ratios are NVIDIA peak figures (H100 SXM). From 1 November 2026, Salad's $0.60 RTX 5090 price brings the lowest-price ratio down to 4.55, below the 4.72 peak ratio. ## When do I need the H100 instead? Rent the H100 when the cheap card runs out of memory or of links between GPUs: - **Models over 32 GB on one GPU:** Llama 3.1 70B in 4-bit (42 GB), QLoRA on 70B (52.5 GB), 8B models with long contexts in a BF16 cache. - **Multi-GPU training:** H100 SXM NVLink runs at 900 GB/s. NVIDIA's RTX 5090 page does not mention NVLink, so plan multi-GPU work over PCIe Gen5. - **Large-batch serving or training on a deadline:** 4.72 times the dense peak math. Rent the RTX 5090 for ComfyUI, SDXL, FLUX, video models that fit in 32 GB, 7B–8B chat models and LoRA runs. See the [best GPU for ComfyUI and Flux](https://gpuwithcrypto.com/best-gpu-for/comfyui-flux/ "What is the best cloud GPU for ComfyUI, Stable Diffusion and Flux?"). For 80 GB at a lower price than the H100, read [H100 vs A100](https://gpuwithcrypto.com/compare/h100-vs-a100/ "H100 vs A100: which should I rent?"). ## Questions about the RTX 5090 and the H100 ### Is the RTX 5090 good for AI? Yes, for work that fits in 32 GB: 7B–8B LLM inference and LoRA, SDXL and FLUX image generation. It has 209.5 dense BF16 TFLOPS, 419 dense FP8 TFLOPS and FP4 Tensor Cores, at about a fifth of the H100's hourly price on 29 October 2026. ### Can an RTX 5090 run Llama 3.1 70B? Not on one card. In 4-bit it needs about 42 GB, so two RTX 5090s; in FP8 about 84 GB, so three. One H100 80GB runs the 4-bit version. ### Is the RTX 5090 faster than the H100? No. NVIDIA's peak figures give the H100 SXM 4.72 times the dense BF16 and FP8 math and 1.87 times the bandwidth. The 5090 has FP4 Tensor Cores, which the H100 lacks, and costs far less per hour. ### How much does it cost to rent an RTX 5090 per hour? On 29 October 2026 the lowest self-serve on-demand price we found was $0.50 (Salad, rising to $0.60 on 1 November) and the median of 3 providers was $0.84. Vast.ai's public price feed showed a lowest marketplace offer of $0.39. GPUwithCrypto charges $0.42/hour. ### Can I rent an RTX 5090 with crypto? Yes. GPUwithCrypto rents it for BTC, USDT, USDC, ETH, SOL, LTC, XMR, with no KYC and no card. The three other providers with an on-demand RTX 5090 price also take crypto: Salad, Spheron, and RunPod, which asks for KYC before the first crypto payment. **Rent an RTX 5090 or an H100 with crypto** RTX 5090 $0.42/hour · H100 $2.32/hour · billed per minute · no KYC. [Rent an RTX 5090](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") · [Rent an H100](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") ## Sources 1. [NVIDIA GeForce RTX 5090](https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/) — NVIDIA (read 2026-10-29) 2. [NVIDIA RTX Blackwell GPU architecture whitepaper](https://images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdf) — NVIDIA (read 2026-10-29) 3. [NVIDIA H100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/h100/) — NVIDIA (read 2026-10-29) 4. [NVIDIA H100 PCIe product brief (PB-11133-001)](https://www.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf) — NVIDIA (read 2026-10-29) 5. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) 6. [The Llama 3 Herd of Models (Table 3)](https://arxiv.org/abs/2407.21783) — arXiv (read 2026-10-29) 7. [FLUX.1 [dev] model card](https://huggingface.co/black-forest-labs/FLUX.1-dev) — Black Forest Labs / Hugging Face (read 2026-10-29) 8. [SDXL 1.0 announcement](https://stability.ai/news-updates/stable-diffusion-sdxl-1-announcement) — Stability AI (read 2026-10-29) 9. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 10. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 11. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # RunPod vs Vast.ai: which is cheaper and more reliable? > RunPod vs Vast.ai: which is cheaper and more reliable? Vast.ai H100 offers start at $1.73 vs RunPod's $1.99; RunPod asks for KYC to pay in crypto. Oct 2026. - URL: https://gpuwithcrypto.com/compare/runpod-vs-vast-ai/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) RunPod and Vast.ai are the two GPU clouds people compare most. Both bill per second and accept crypto. They differ in how prices are set, who runs the machines, and what happens when you pay in crypto. **Quick answer — RunPod vs Vast.ai: which is cheaper and more reliable?** **RunPod vs Vast.ai:** Vast.ai is cheaper; RunPod sells fixed list prices. As of 29 October 2026, Vast.ai's public price feed showed an H100 SXM from $1.73 per hour, against $3.99 on RunPod's Secure Cloud and $2.69 on its Community Cloud. On reliability the difference is the model: Vast.ai is a marketplace of hosts, RunPod sells two fixed-price tiers. Both take crypto; RunPod asks for KYC first. **GPUwithCrypto** adds on-demand GPUs for crypto, no KYC. ## How do RunPod and Vast.ai differ at a glance? From each provider's pricing page, billing docs and terms, read on 29 October 2026. GPUwithCrypto publishes this page and is shown as a third column. | | RunPod | Vast.ai | GPUwithCrypto | | --- | --- | --- | --- | | Model | Secure Cloud at RunPod list prices; Community Cloud on community-hosted machines | Marketplace: hosts list offers, price and hardware vary by offer | On-demand whole GPUs at one hourly price per model | | H100 from, per GPU-hour | $2.89 PCIe, $3.99 SXM (Secure); $1.99 PCIe, $2.69 SXM (Community) | $1.73 H100 SXM (lowest offer) | $2.32/hour | | Billing | Per second | Per second | Per minute | | Footprint | 30+ regions | 40+ data centers | Host country shown before you start | | Longer terms | Reserved pricing via sales | 1-, 3- and 6-month reserved terms, 'up to 50% off' | Prepaid blocks of GPU-hours at the same rate | | Pays in crypto | Yes; Crypto.com named in its terms, coins not listed | Yes, via BitPay and Crypto.com | BTC, USDT, USDC, ETH, SOL, LTC, XMR | | ID / KYC | KYC before the first crypto payment | No ID step in its billing docs, FAQ or terms | None | | Card | Auto-pay is card-only; prepaid cards need $100 per deposit | Billing FAQ: credit is added after a credit card and an email are verified | No card | | Minimum to start | $10 | Not stated | $1 | | Refunds | Credits non-refundable and non-withdrawable | No refund on spent credits; credits bought with crypto are not refunded | Hours of a failed instance go back to your balance; balance never expires | | Spend limit | $80 per hour by default, raised over time | — | — | *Prices checked on October 29, 2026.* 'No ID step' means none is described in the pages we read, not a promise. Payment processors such as BitPay and Crypto.com can run their own checks. ## How do RunPod and Vast.ai prices compare per GPU? Per GPU-hour, USD, read on 29 October 2026. RunPod prices come from its pricing page (header 'Updated October 8, 2026'). Vast.ai figures come from Vast.ai's public price feed (updated 29 October 2026, 10:30 UTC): the cheapest rentable offer and the median offer, which mix on-demand and interruptible listings. | GPU | RunPod Secure Cloud | RunPod Community Cloud | Vast.ai lowest offer | Vast.ai median offer | GPUwithCrypto | | --- | --- | --- | --- | --- | --- | | H100 SXM 80GB | $3.99 | $2.69 | $1.73 | $2.40 | $2.32/hour (H100 80GB) | | H100 PCIe 80GB | $2.89 | $1.99 | $2.27 (1 offer) | — | $2.32/hour (H100 80GB) | | H200 141GB | $5.29 | $3.59 | $3.00 (NVL); $3.77 | $4.00 (NVL); $5.05 | $3.07/hour | | B200 | $7.99 | $5.98 | $7.50 | $8.75 | $4.61/hour | | A100 80GB | $1.79 | $1.19 PCIe, $1.39 SXM | $0.34 PCIe, $0.48 SXM | $0.68 PCIe, $0.80 SXM | $0.92/hour | | L40S 48GB | $1.09 | $0.79 | $0.33 | $0.73 | $0.82/hour | | RTX 4090 | $0.89 | $0.34 | $0.13 | $0.49 | $0.28/hour | | RTX 5090 | $1.19 | $0.69 | $0.39 | $0.73 | $0.42/hour | *Prices checked on October 29, 2026.* RunPod lists one A100 80GB Secure Cloud price for PCIe and SXM. Vast.ai figures are displayed as published in Vast.ai's public price feed, with attribution. GPUwithCrypto prices are on-demand, each at least 15% under a named on-demand reference read on 29 October 2026; GPUwithCrypto H100 hosts are SXM or PCIe, shown before you start. ## Which is cheaper, RunPod or Vast.ai? Vast.ai, on its lowest offers. Its cheapest rentable offer was below RunPod's Community Cloud price for 6 of the 8 GPUs in the table; the exceptions were the H100 PCIe and the B200. Against RunPod's Secure Cloud, Vast.ai's lowest offer was cheaper for every GPU. The median offer tells a closer story. Vast.ai's median was still below RunPod's Community Cloud for the H100 SXM, the A100 and the L40S. RunPod's Community Cloud was cheaper than Vast.ai's median for the H200, the B200, the RTX 4090 and the RTX 5090. According to GPUwithCrypto's price check of 29 October 2026, RunPod's $2.89 Secure Cloud H100 PCIe was the second-lowest self-serve on-demand H100 price among 16 providers. Only Massed Compute ($2.73) was lower. The full ranking is on the [RunPod alternatives](https://gpuwithcrypto.com/alternatives/runpod/ "What are the best (cheaper) RunPod alternatives?") page. ## Which is more reliable, RunPod or Vast.ai? This page does not rank them on reliability: our research covers prices, billing, crypto and ID rules, not measured uptime. What differs is the model: - **RunPod Secure Cloud** is RunPod's on-demand tier at a fixed list price. **Community Cloud** runs on community-hosted machines at a lower price. - **Vast.ai** is a marketplace. Hosts list offers, and price and hardware vary by offer. Its price feed does not separate on-demand from interruptible offers, so check the offer type before a long job. - **Both** bill GPU time per second. RunPod still charges for volume storage while a pod is stopped ($0.20 per GB per month idle, $0.10 while running). If you search for Vast.ai reviews, note that news about VAST Data concerns a different company. Interruptible capacity is explained in [what is a spot GPU](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?"). ## Do RunPod and Vast.ai accept crypto without KYC? Both accept crypto, and only RunPod documents a KYC step for it. RunPod's billing docs say to complete any required KYC verification before your first crypto payment, and its auto-pay works with cards only. Its terms name Crypto.com as a payment provider; the coins are not listed. Vast.ai takes crypto through BitPay and Crypto.com, with Coinbase Commerce referenced as a legacy option. It does not refund credits bought with crypto. Its billing FAQ says credit is added after a credit card and an email are verified; the docs do not say whether crypto-only top-ups skip the card. Full answers: [Does RunPod accept crypto?](https://gpuwithcrypto.com/does-runpod-accept-crypto/ "Does RunPod accept crypto?") and [Does Vast.ai accept crypto?](https://gpuwithcrypto.com/does-vast-ai-accept-crypto/ "Does Vast.ai accept crypto?"). ## What is each one good for? **RunPod** suits teams that want a fixed list price, a choice between Secure and Community Cloud, and per-second billing across 30+ regions. Card payers get auto-pay. It sells B200s on-demand at $7.99 (Secure) and $5.98 (Community). **Vast.ai** suits price hunters who can pick and vet an offer. It had the lowest offers for most GPUs and the widest consumer-card choice: 970 rentable RTX 5090 offers and 562 RTX 4090 offers in its feed on 29 October 2026. Reserved terms of 1 to 6 months go up to 50% off. Both suit card payers best: RunPod's auto-pay is card-only, and Vast.ai's billing FAQ asks for a verified card. ## Where does GPUwithCrypto fit between RunPod and Vast.ai? **GPUwithCrypto** is a third option for one profile: on-demand GPUs, paid in crypto, with no KYC and no card. Each GPU is priced at least 15% under a named reference: the on-demand price of the same GPU at a self-serve provider, read on 29 October 2026 (H100: Massed Compute at $2.73 per GPU-hour; each GPU's reference is named on its price page). Spot, interruptible, marketplace and contact-sales prices are not used as references, and some of them are lower. We do not claim to be the cheapest for every GPU. - **Against RunPod:** our price is below RunPod's Secure Cloud for every GPU in the table. RunPod's Community Cloud is lower than ours for the H100 PCIe and the L40S. - **Against Vast.ai:** Vast.ai's lowest offer is lower than ours for every GPU in the table except the B200. We sell one fixed on-demand price, not a marketplace offer. - **Payment:** BTC, USDT, USDC, ETH, SOL, LTC, XMR, from $1, with no ID, no selfie and no card. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. See every price on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"), or compare the [GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?"). ## Questions about RunPod and Vast.ai ### Is Vast.ai cheaper than RunPod? On its lowest offers, yes: on 29 October 2026 Vast.ai's public price feed showed an H100 SXM from $1.73, against $3.99 on RunPod Secure Cloud and $2.69 on Community Cloud. At the median offer, RunPod's Community Cloud was cheaper for the H200, B200, RTX 4090 and RTX 5090. ### What is the difference between RunPod Secure Cloud and Community Cloud? Secure Cloud is RunPod's on-demand tier at its list price; Community Cloud runs on community-hosted machines at a lower price. For an H100 SXM on 29 October 2026 that was $3.99 against $2.69 per hour. ### Does RunPod require KYC to pay with crypto? Yes, according to its billing docs read on 29 October 2026: complete any required KYC verification before your first crypto payment. Its auto-pay works with cards only. GPUwithCrypto takes crypto with no KYC. ### Does Vast.ai refund crypto payments? No. Vast.ai's billing reference says it does not refund credits bought with crypto, and spent credits are never refunded. Unspent credits bought another way can be refunded at Vast.ai's discretion, per its terms. ### Do RunPod and Vast.ai bill per second? Yes, both bill GPU time per second. RunPod also charges for volume storage while a pod is stopped. GPUwithCrypto bills per minute while an instance runs. ## Sources 1. [RunPod pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 2. [RunPod billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 3. [RunPod terms of service](https://www.runpod.io/legal/terms-of-service) — RunPod (read 2026-10-29) 4. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 5. [Vast.ai pricing](https://vast.ai/pricing) — Vast.ai (read 2026-10-29) 6. [Vast.ai billing reference](https://docs.vast.ai/documentation/reference/billing) — Vast.ai (read 2026-10-29) 7. [Vast.ai billing FAQ](https://docs.vast.ai/guides/reference/faq/billing.md) — Vast.ai (read 2026-10-29) 8. [Vast.ai terms of service](https://vast.ai/terms) — Vast.ai (read 2026-10-29) 9. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How do I contact GPUwithCrypto? > Email GPUwithCrypto support about a GPU instance, a crypto payment or a refund, and report misuse to the abuse address. What to include, and what never to send. - URL: https://gpuwithcrypto.com/contact/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Support works by email. Write to the right address, include the details below, and never send a private key, an API key or your account number. **Quick answer — How do I contact GPUwithCrypto?** Email support@gpuwithcrypto.com for help with a **GPUwithCrypto** GPU instance, a crypto payment or your account, and abuse@gpuwithcrypto.com to report misuse of a GPU. A person answers support emails within 12 hours, seven days a week, and helps you start, size and stop your GPU instances. As of 29 October 2026 support works by email only, with no phone line or live chat. Never send your SSH private key, an API key or your account number: support does not need them. ## Contact details | Attribute | Value | | --- | --- | | Support | support@gpuwithcrypto.com | | Abuse reports | abuse@gpuwithcrypto.com | | Reply time | A person replies within 12 hours | | Language | English | | Phone or live chat | None: email only | ## What should a support email contain? Enough to find the instance or the payment without a back-and-forth: - **For an instance problem:** the instance ID, the GPU model, the time in UTC, what you ran and the exact error. For a GPU fault, paste the output of `nvidia-smi`. - **For a payment:** the coin, the network, the amount sent and the transaction ID. - **For a refund:** which case of the [refund policy](https://gpuwithcrypto.com/refunds/ "Refund policy") applies, with the instance ID or the transaction ID. An instance ID or a transaction ID is enough for support to find your records. Do not include your account number, an API key or a private key. ## What can support help with? Support helps you start, size and stop GPU instances: - Choosing a GPU for a model and its VRAM needs, with the [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") as a starting point. - Connecting over SSH and starting from your own Docker image. - Instances that fail to start, and GPUs that do not work. - Top-ups that arrive late or with the wrong amount. - Refund requests under the refund policy. - Questions about what the [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") allows, before you start a job. Support cannot read or recover your files, because we do not read your disk. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. Support cannot recover a lost account number without an email on file either. ## How do I report abuse coming from a GPUwithCrypto GPU? Write to abuse@gpuwithcrypto.com. A useful report gives: - the IP address the abuse came from; - the date and time, with the time zone; - what happened: attack traffic, scanning, spam, malware, or content you believe is illegal; - the evidence you can share, such as log lines or message headers; - how to reach you if we need more detail. Never attach child sexual abuse material to an email: describe it and say where it is. Reports go to abuse@gpuwithcrypto.com. An instance that breaks the acceptable use policy is stopped and the hours it used are not refunded; child sexual abuse material is reported to the authorities. What counts as misuse is listed in the [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy"). ## How do I raise a payment problem? Write to support@gpuwithcrypto.com with the transaction ID, the coin, the network and the amount. A top-up is credited after the network confirmations, so a busy network delays it without losing it. Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited. If you sent a coin or used a network that the payment page did not show, write before sending anything else: such funds are not always recoverable. ## How can I tell a real message from GPUwithCrypto? By what it asks for. Nobody from our team will ever ask for your account number, your SSH private key or your API keys, by email or in a chat. Our only addresses are the two on this page. If a message asks for any of these, it is not from us: do not answer it, and forward it to abuse@gpuwithcrypto.com. ## Questions about support ### How fast does support reply? A person answers support emails within 12 hours, seven days a week, and helps you start, size and stop your GPU instances. Email keeps a written record of each request, which matters when a payment or a refund is involved. Put the instance ID or the transaction ID in the first message to get a complete answer in one reply. ### Can support recover my account number? Only if you added an email address. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good. Store the number in a password manager when you open the account. ### Can support get back the files of a deleted instance? No. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. Nobody can restore a wiped disk, so copy checkpoints and outputs to your own storage before you delete an instance. ### Do I have to give my name to get help? No. Support needs the instance ID or the transaction ID, not your name, a document or a card. Sign in with an account number; an email address is optional. Sign a support email however you like: what matters is the ID that lets support find the instance or the payment. --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Does RunPod accept crypto? > Does RunPod accept crypto? Yes, via Crypto.com, but it asks for KYC before the first crypto payment and credits are non-refundable. Checked on 29 Oct 2026. - URL: https://gpuwithcrypto.com/does-runpod-accept-crypto/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) RunPod takes crypto, but not on the same terms as a card: KYC first, no auto-pay, no refund. Here is what its own pages say, and which clouds take crypto without an ID check. **Quick answer — Does RunPod accept crypto?** Yes, RunPod accepts crypto, but it asks for KYC verification before your first crypto payment. As of 29 October 2026, its billing docs say crypto is accepted through integrated payment processors, and its Terms name Crypto.com. Auto-pay is card-only, new users can start with $10, and credits are non-refundable. **GPUwithCrypto** rents an H100 for $2.32/hour in crypto with no ID check. ## What do RunPod's own pages say about crypto? Six facts from RunPod's billing documentation and Terms of Service, both read on 29 October 2026. | Question | What RunPod says | Source | | --- | --- | --- | | Which coins? | Not named. Crypto is "accepted via integrated payment processors". | [Billing docs](https://docs.runpod.io/references/billing-information) | | Which processor? | Crypto.com, named in Terms section 8 as a third-party payment provider; cards go through Stripe | [Terms of Service](https://www.runpod.io/legal/terms-of-service) | | ID or KYC? | "Complete any required KYC verification before your first crypto payment." | [Billing docs](https://docs.runpod.io/references/billing-information) | | Auto-pay? | Card only: automatic payments do not run on crypto | [Billing docs](https://docs.runpod.io/references/billing-information) | | Minimum? | New users "can start with as little as $10"; prepaid cards need at least $100 per transaction | [Billing docs](https://docs.runpod.io/references/billing-information) | | Refunds? | Credits are non-refundable and cannot be withdrawn once deposited | [Billing docs](https://docs.runpod.io/references/billing-information) | *Prices checked on October 29, 2026.* ## Why does RunPod ask for KYC before a crypto payment? RunPod's docs state the rule but not the reason. The KYC note sits next to crypto only: card payments go through Stripe and carry no such step. According to GPUwithCrypto's check of 29 October 2026, 14 of 33 GPU clouds accept crypto. RunPod is the only one of the 14 whose docs tie KYC to a crypto payment. Business customers spending more than $5,000 can also be invoiced by ACH, wire or card through RunPod's sales team. In practice, paying RunPod in crypto means an identity check first, then a top-up you cannot get back. Plan the first deposit around the job you have. ## How much does RunPod cost compared with GPUwithCrypto? RunPod lists two prices per GPU: Secure Cloud and Community Cloud. Community Cloud runs on community-hosted machines; RunPod does not describe it as interruptible spot. GPUwithCrypto prices are on-demand, per GPU-hour, paid in crypto. | GPU | RunPod Secure Cloud | RunPod Community Cloud | GPUwithCrypto | | --- | --- | --- | --- | | H100 80GB | $2.89 (PCIe), $3.99 (SXM) | $1.99 (PCIe), $2.69 (SXM) | **$2.32/hour** | | H200 141GB | $5.29 | $3.59 | **$3.07/hour** | | B200 | $7.99 | $5.98 | **$4.61/hour** | | A100 80GB | $1.79 (PCIe or SXM) | $1.19 (PCIe), $1.39 (SXM) | **$0.92/hour** | | L40S | $1.09 | $0.79 | **$0.82/hour** | | RTX 4090 | $0.89 | $0.34 | **$0.28/hour** | | RTX 5090 | $1.19 | $0.69 | **$0.42/hour** | *Prices checked on October 29, 2026.* USD per GPU-hour, from runpod.io/pricing (page header "Updated October 8, 2026"), read on 29 October 2026. RunPod bills per second. On 29 October 2026, GPUwithCrypto was cheaper than RunPod Secure Cloud for all seven GPUs. RunPod Community Cloud was cheaper than GPUwithCrypto for the H100 PCIe and the L40S, and dearer for the other five. RunPod also lists GPU types that GPUwithCrypto does not rent, such as the L4, L40 and RTX A6000. ## What else should I know before paying RunPod in crypto? Four more rules from RunPod's own pages: - **Spend limit:** new accounts have a default limit of $80 per hour across all resources, raised over time or through support. - **Countries:** its Terms (section 6) name Cuba, Iran, North Korea and Syria as embargoed. - **Billing:** pods are billed per second, and storage is extra: container disk $0.10 per GB per month, volume disk $0.20 per GB per month while idle. - **Regions:** its pricing page lists 30+ regions. RunPod's strengths are real: per-second billing, a $10 start, and on-demand prices for all seven GPUs we track, which none of the other 13 crypto-accepting clouds in our check listed on 29 October 2026. ## Which GPU clouds take crypto without asking for ID? Several, according to their own pages on 29 October 2026. "No ID step documented" means the pages we read describe none, not a promise that they never ask. - **GPUwithCrypto**: BTC, USDT, USDC, ETH, SOL, LTC, XMR, no ID, no card. H100 $2.32/hour, on-demand only, billed per minute. - **Massed Compute**: Bitcoin by wallet invoice, email verification only. H100 80GB $2.73 on-demand, but no auto-recharge for crypto. - **Vast.ai**: BitPay and Crypto.com, no ID document in its docs. Its billing FAQ asks for a verified card and email before credit is added, and crypto credits are not refunded. Details on [the Vast.ai crypto page](https://gpuwithcrypto.com/does-vast-ai-accept-crypto/ "Does Vast.ai accept crypto?"). - **Salad**: USDC, USDT and RENDER on Solana, for RTX 4090 and RTX 5090. Crypto payments cannot be refunded. - **Spheron**: USDC on six chains, from a $50 crypto deposit. - **Vultr**: nine coins through BitPay, for L40S and A100 plans. The [full table of 33 GPU clouds](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") adds processors, minimums and refund rules. For other trade-offs, see [RunPod alternatives](https://gpuwithcrypto.com/alternatives/runpod/ "What are the best (cheaper) RunPod alternatives?"). ## RunPod and crypto: questions ### Can I pay RunPod with Bitcoin or USDT? RunPod's docs do not name the coins. They say crypto is accepted through integrated payment processors, and the Terms name Crypto.com. Either way, KYC comes before the first crypto payment. ### Why is my RunPod payment not working? Check three documented rules. Prepaid cards need at least $100 per transaction, crypto needs KYC first, and new accounts have an $80 per hour spend limit across all resources. ### Can I use a prepaid card on RunPod? Yes, with a minimum of $100 per transaction, per its billing docs. New users can otherwise start with $10. ### Can I get my RunPod credits back? No. RunPod's billing docs say credits are non-refundable and cannot be withdrawn once deposited, whatever the payment method. ### Is there a RunPod alternative that takes crypto without KYC? Yes. GPUwithCrypto, Massed Compute, Salad, Spheron and Vultr take crypto and describe no ID step. GPUwithCrypto rents an H100 for $2.32/hour with BTC, USDT, USDC, ETH, SOL, LTC, XMR. ## Sources 1. [Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 2. [Terms of service](https://www.runpod.io/legal/terms-of-service) — RunPod (read 2026-10-29) 3. [Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 4. [Crypto: manually add credits](https://vm-docs.massedcompute.com/docs/billing/crypto/manually-add-credits) — Massed Compute (read 2026-10-29) 5. [Billing reference](https://docs.vast.ai/documentation/reference/billing) — Vast.ai (read 2026-10-29) 6. [Billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 7. [Billing](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 8. [What payment methods do you accept?](https://docs.vultr.com/support/platform/billing/what-payment-methods-do-you-accept) — Vultr (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Does Vast.ai accept crypto? > Does Vast.ai accept crypto? Yes, through BitPay and Crypto.com, but credits bought with crypto are not refunded. Rules, prices and no-ID options, 29 Oct 2026. - URL: https://gpuwithcrypto.com/does-vast-ai-accept-crypto/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Vast.ai takes crypto through two payment processors. What its docs add matters as much: a card and an email to verify, and no refund on crypto credits. **Quick answer — Does Vast.ai accept crypto?** Yes, Vast.ai accepts crypto through BitPay and Crypto.com, but it does not refund credits bought with crypto. As of 29 October 2026, its docs name no coins and no minimum, and its billing FAQ says credit is added after a credit card and an email are verified. **GPUwithCrypto** takes crypto with no card and no ID: an H100 costs $2.32/hour. ## What do Vast.ai's own pages say about crypto? Six facts from Vast.ai's billing reference, billing FAQ and Terms (version dated 1 September 2026), all read on 29 October 2026. | Question | What Vast.ai says | Source | | --- | --- | --- | | Which coins? | Not named: the docs name the processors only | [Billing reference](https://docs.vast.ai/documentation/reference/billing) | | Which processor? | BitPay and Crypto.com; Coinbase Commerce is referenced as a legacy option | [Billing reference](https://docs.vast.ai/documentation/reference/billing) | | ID or KYC? | No identity-document requirement in the billing docs, billing FAQ or Terms | [Terms](https://vast.ai/terms) | | Card needed? | Credit is added "after you enter a credit card and an email address and both are verified"; the docs do not say if crypto-only top-ups are exempt | [Billing FAQ](https://docs.vast.ai/guides/reference/faq/billing.md) | | Minimum? | No minimum deposit stated | [Billing FAQ](https://docs.vast.ai/guides/reference/faq/billing.md) | | Refunds? | "We do not refund Vast credits bought with crypto"; Coinbase Commerce payments cannot be refunded either | [Billing reference](https://docs.vast.ai/documentation/reference/billing) | *Prices checked on October 29, 2026.* ## What happens to Vast.ai credits I do not use? If you paid in crypto, you cannot get them back as money. Spent credits are never refunded, and its Terms leave a refund of unspent credits to its discretion, depending on the payment method. Credits bought with crypto are excluded outright. That makes the first top-up a sizing decision. Vast.ai bills per second with no minimum hours, so a small deposit is enough to test a host before a long run. ## How much does Vast.ai cost compared with GPUwithCrypto? Vast.ai is a marketplace, so the price depends on the host. On 29 October 2026 the Vast.ai public price feed listed an H100 SXM from $1.73 per hour, with a median offer of $2.40 and 56 offers rentable. The feed does not split on-demand from interruptible offers. Vast.ai also sells 1-, 3- and 6-month reserved terms, advertised at "Up to 50% Off". GPUwithCrypto rents an on-demand H100 for $2.32/hour, billed per minute. Vast.ai's cheapest H100 offer was below that price; its median offer was above it. The choice is between the lowest marketplace offer and a fixed on-demand price: Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. For a two-provider view, see [RunPod vs Vast.ai](https://gpuwithcrypto.com/compare/runpod-vs-vast-ai/ "RunPod vs Vast.ai: which is cheaper and more reliable?"). ## What else does Vast.ai require? Its Terms add three conditions: - **Currency:** all payments are in US dollars, whatever you pay with. - **Location:** users represent that they are not located in or resident of Cuba, Iran, North Korea, Syria or other restricted regions. - **Export controls:** Vast.ai can suspend service over export-control concerns. Vast.ai's strengths are real: per-second billing, no minimum hours, 40+ data centers on its pricing page, and the lowest marketplace H100 offer we found on 29 October 2026. ## Which GPU clouds take crypto without a card or an ID check? A few, according to their own pages on 29 October 2026. According to GPUwithCrypto's check of that day, 14 of 33 GPU clouds accept crypto. Twelve of the 14 describe no ID step, Vast.ai included, but Vast.ai still asks for a verified card. "No ID step documented" means the pages we read describe none, not a promise that they never ask. - **GPUwithCrypto**: BTC, USDT, USDC, ETH, SOL, LTC, XMR, no ID and no card at any step. H100 $2.32/hour, A100 80GB $0.92/hour, RTX 4090 $0.28/hour. - **Massed Compute**: Bitcoin by wallet invoice, email verification only. H100 80GB $2.73 on-demand; no auto-recharge for crypto. - **Salad**: USDC, USDT and RENDER on Solana, for RTX 4090 and RTX 5090. Crypto payments cannot be refunded. - **Spheron**: USDC on six chains, from a $50 crypto deposit; sign-up needs a verified email. - **Akash Network**: Console Air needs only a wallet and AKT, with no account. RunPod takes crypto too, after KYC: see [does RunPod accept crypto](https://gpuwithcrypto.com/does-runpod-accept-crypto/ "Does RunPod accept crypto?"). The [full table of 33 GPU clouds](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") and [Vast.ai alternatives](https://gpuwithcrypto.com/alternatives/vast-ai/ "What are cheaper or more reliable Vast.ai alternatives?") cover the rest. ## Vast.ai and crypto: questions ### Can I pay Vast.ai with Bitcoin or USDT? Vast.ai's docs do not name the coins. Crypto goes through BitPay or Crypto.com, and the docs list no coins or networks. ### Does Vast.ai need a credit card? Its billing FAQ says credit is added once a credit card and an email address are both verified. The docs do not say whether a crypto-only top-up skips the card step. ### Why is my Vast.ai payment not working? Start with the two checks its billing FAQ describes: a verified email and a verified credit card. Payments are in US dollars under its Terms, and some regions are excluded. ### Can I get a refund on Vast.ai credits bought with crypto? No. Vast.ai's billing reference says: "We do not refund Vast credits bought with crypto." Coinbase Commerce payments cannot be refunded either. ### Is there a Vast.ai alternative that takes crypto with no card? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no card and no ID, from $1, and rents on-demand GPUs from $0.28/hour. ## Sources 1. [Billing reference](https://docs.vast.ai/documentation/reference/billing) — Vast.ai (read 2026-10-29) 2. [Billing FAQ](https://docs.vast.ai/guides/reference/faq/billing.md) — Vast.ai (read 2026-10-29) 3. [Terms of service](https://vast.ai/terms) — Vast.ai (read 2026-10-29) 4. [Public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 5. [Crypto: manually add credits](https://vm-docs.massedcompute.com/docs/billing/crypto/manually-add-credits) — Massed Compute (read 2026-10-29) 6. [Billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 7. [Billing](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 8. [Console Air](https://akash.network/docs/developers/deployment/console-air) — Akash Network (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # GPU rental with crypto: frequently asked questions > Renting a GPU with crypto: H100 price per hour, no KYC, coins, minimum top-up, per-minute billing, hour blocks, refunds and SSH. Updated 29 October 2026. - URL: https://gpuwithcrypto.com/faq/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Short answers to the questions people ask before renting a GPU with crypto. Each one links to the page with the details. **Quick answer — What should I know before renting a GPU with crypto?** **GPUwithCrypto** rents on-demand NVIDIA GPUs billed per minute from GPU-hours bought in crypto, with no KYC: no ID, no selfie, no card. As of 29 October 2026 an H100 80GB costs $2.32/hour and an RTX 4090 $0.28/hour, each at least 15% under a named on-demand reference. The minimum top-up is $1, and balance never expires. ## Price and billing ### How much does it cost to rent an H100 per hour? An NVIDIA H100 80GB costs $2.32/hour on GPUwithCrypto as of 29 October 2026, billed per minute. According to GPUwithCrypto's price check of that day, the lowest self-serve on-demand price elsewhere was $2.73 at Massed Compute, and the median across 16 providers was $4.16. An H200 costs $3.07/hour, an A100 80GB $0.92/hour and an RTX 4090 $0.28/hour. The [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?") lists every GPU, and the [price index](https://gpuwithcrypto.com/gpu-price-index/ "GPU rental price index: what did a GPU-hour cost in October 2026?") shows each reference price. ### Is GPU rental billed per minute or per hour? Per minute, at the hourly rate. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. A 90-minute job costs one and a half times the hourly price. An instance you forget to stop keeps using GPU-hours until you stop it or its GPU-hours run out. ### What is a block of GPU-hours? Prepaid GPU time for one GPU model. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. Buying a block does not start an instance: you start one from the dashboard when that GPU is free. A 24-hour H100 block costs $55.68 and a 10-hour A100 block $9.20. ### Is there a free trial? No. There is no free tier: a 10-hour block is the smallest purchase to try a GPU, and unused hours never expire. The smallest purchase is a 10-hour block; on the RTX 4090 it costs $2.80. The minimum top-up is only $1, so you can top up that exact block price and nothing more. ### Can I get a refund? In defined cases. If an instance fails to start, or its GPU does not work and support cannot fix it, the hours it used go back to your balance. Unused GPU-hours and balance never expire. Top-ups are not paid out unless the law requires it; the part added only to reach the minimum top-up comes back on request within 30 days of payment, in USDT or USDC, as long as it is unspent. The [refund policy](https://gpuwithcrypto.com/refunds/ "Refund policy") works through examples. ## Crypto and no KYC ### Can I rent a GPU and pay with crypto? Yes: crypto is the only way to pay on GPUwithCrypto. You top up a balance in BTC, USDT, USDC, ETH, SOL, LTC, XMR, and it is credited after the network confirmations. There is no card, PayPal or bank transfer option. According to GPUwithCrypto's check of 29 October 2026, 14 of the 33 GPU clouds we checked accept crypto; the [comparison of GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") lists them with their coins. ### Is there a GPU cloud with no KYC? Yes: GPUwithCrypto asks for no ID, no selfie, no documents and no card. Sign in with an account number; an email address is optional. Of the 33 GPU clouds we checked on 29 October 2026, 4 ask for ID in some cases, and RunPod asks for KYC before the first crypto payment. The rules still apply to every account. Details on the [no-KYC GPU cloud](https://gpuwithcrypto.com/no-kyc-gpu-cloud/ "Is there a GPU cloud with no KYC?") page. ### What is the minimum top-up? The minimum top-up is $1, in any accepted coin. Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. ## Instances, data and API ### Are GPUs available right now? It depends on the GPU and the hour. Stock changes through the day: the dashboard shows which GPUs are free right now, and nothing is charged until an instance starts. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. Unused balance and GPU-hours never expire, so nothing is lost while you wait for a busy model. ### Do I get root SSH access, and can I use my own Docker image? Yes. Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. You add your SSH public key in the dashboard; we never see or store your private key. The [step-by-step guide](https://gpuwithcrypto.com/guides/rent-a-gpu-for-ai/ "How do I rent a GPU for AI, step by step?") shows how to connect and run a first job. ### What happens to the data on my instance's disk? It is yours: we do not read the files on your disk. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. When the GPU-hours of a GPU run out, its instances stop and their disks are kept; a new block of hours lets you start them again. Keep a margin of GPU-hours during long runs and copy checkpoints out regularly. ### Is there an API to start and stop GPUs? Yes. A REST API starts, stops and lists instances with an API key, so a script or an agent can run jobs without the dashboard. It uses the same balance, the same GPUs and the same per-minute rate as the dashboard, so a training script can start an instance, run its job and stop it. ## Rules and countries ### Which countries can rent GPUs from GPUwithCrypto? Most countries, with exceptions set by sanctions and export rules. Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine. Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited. ### What can I not do with a rented GPU? No cryptocurrency mining, no password or hash cracking, no attacks on other systems (DDoS, scanning, exploitation), no malware, no spam, no child sexual abuse material, no intimate or deceptive images of real people without their consent, no weapons development and no work for sanctioned people or companies. The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") explains each rule. An instance that breaks one is stopped, and the hours it used are not refunded. ## Sources 1. [Massed Compute pricing (H100 80GB reference price)](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Billing information (KYC before the first crypto payment)](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # GPU rental price index: what did a GPU-hour cost in October 2026? > GPU rental price index, October 2026: the median on-demand H100 cost $4.16 per GPU-hour across 16 providers. Six more GPUs, free, from a crypto GPU cloud. - URL: https://gpuwithcrypto.com/gpu-price-index/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) A monthly snapshot of self-serve on-demand GPU prices: lowest, 25th percentile, median and highest price per GPU-hour, for seven NVIDIA GPUs. This is the first edition, so there is no earlier month to compare yet. **Quick answer — What is the GPU rental price index for October 2026?** **GPUwithCrypto**'s GPU rental price index for October 2026 is a first edition, with no earlier month to compare. On 29 October 2026 the median on-demand H100 80GB rented for $4.16 per GPU-hour across 16 providers, the H200 for $5.29 across 13 and the A100 80GB for $2.23 across 13. GPUwithCrypto, which rents GPUs for crypto, prices each GPU at least 15% under a named on-demand reference. ## What did each GPU rent for on 29 October 2026? Self-serve on-demand list prices in USD per GPU-hour, one price per provider: its lowest on-demand offer for that GPU. Spot, preemptible and marketplace prices are not in the index. | GPU | Providers | Lowest | 25th percentile | Median | Highest | | --- | --- | --- | --- | --- | --- | | NVIDIA H100 80GB | 16 | $2.73 (Massed Compute) | $3.20 | **$4.16** | $12.29 (Microsoft Azure) | | NVIDIA H200 141GB | 13 | $3.62 (Massed Compute) | $4.47 | **$5.29** | $10.60 (Google Cloud) | | NVIDIA B200 | 9 | $5.43 (Massed Compute) | $6.69 | **$8.50** | $14.24 (AWS) | | NVIDIA A100 80GB | 13 | $1.09 (Thunder Compute) | $1.69 | **$2.23** | $5.07 (Google Cloud) | | NVIDIA L40S 48GB | 11 | $0.97 (Massed Compute) | $1.53 | **$1.67** | $3.15 (Spheron) | | NVIDIA RTX 4090 | 2 | $0.33 (Salad) | — | **$0.61** | $0.89 (RunPod) | | NVIDIA RTX 5090 | 3 | $0.50 (Salad) | — | **$0.84** | $1.19 (RunPod) | *Prices checked on October 29, 2026.* Rounded to the cent; statistics are computed on unrounded prices. No 25th percentile is given below four providers. H100 includes SXM, PCIe, NVL and unstated form factors; H200 includes SXM and NVL. ## What are the key figures of the October 2026 index? One sentence per GPU, each true on 29 October 2026, according to GPUwithCrypto's price check: - The median on-demand **H100 80GB** rented for $4.16 per GPU-hour across 16 providers, from $2.73 to $12.29. - The median on-demand **H200 141GB** rented for $5.29 per GPU-hour across 13 providers, from $3.62 to $10.60. - The median on-demand **B200** rented for $8.50 per GPU-hour across 9 providers, from $5.43 to $14.24. - The median on-demand **A100 80GB** rented for $2.23 per GPU-hour across 13 providers, from $1.09 to $5.07. - The median on-demand **L40S** rented for $1.67 per GPU-hour across 11 providers, from $0.97 to $3.15. - The median on-demand **RTX 4090** rented for $0.61 per GPU-hour across 2 providers, from $0.33 to $0.89. - The median on-demand **RTX 5090** rented for $0.84 per GPU-hour across 3 providers, from $0.50 to $1.19. The spread is wide. The dearest on-demand H100, an 8-GPU Azure VM, cost 4.5 times the cheapest, Massed Compute's single H100. ## Which providers are in the index, and at what price? The 18 providers behind the index and their lowest self-serve on-demand price per GPU-hour on 29 October 2026. "—" means the provider listed no on-demand price for that GPU. | Provider | H100 | H200 | B200 | A100 80GB | L40S | RTX 4090 | RTX 5090 | | --- | --- | --- | --- | --- | --- | --- | --- | | Massed Compute | $2.73 | $3.62 | $5.43 | $1.35 | $0.97 | — | — | | RunPod | $2.89 | $5.29 | $7.99 | $1.79 | $1.09 | $0.89 | $1.19 | | OVHcloud | $2.99 | $6.20 | — | $3.07 | $1.80 | — | — | | Hyperbolic | $3.19 | $3.99 | $5.99 | — | — | — | — | | Thunder Compute | $3.20 | — | — | $1.09 | — | — | — | | Lambda | $3.29 | — | $6.69 | $2.79 | — | — | — | | Jarvislabs | $3.49 | $4.59 | — | $1.49 | — | — | — | | Crusoe | $3.90 | $4.29 | — | $2.00 | $1.50 | — | — | | Paperspace / DigitalOcean | $4.41 | $4.47 | — | — | $1.57 | — | — | | Nebius | $4.50 | $5.40 | $8.50 | — | $1.55 | — | — | | Verda (formerly DataCrunch) | $4.75 | $6.19 | $8.78 | $2.23 | $1.88 | — | — | | Spheron | $5.86 | $4.80 | $11.12 | $1.69 | $3.15 | — | $0.84 | | CoreWeave | $6.16 | $6.31 | $8.60 | $2.70 | $2.25 | — | — | | AWS | $6.88 | $7.91 | $14.24 | $3.43 | $1.86 | — | — | | Google Cloud | $11.06 | $10.60 | — | $5.07 | — | — | — | | Microsoft Azure | $12.29 | — | — | — | — | — | — | | Vultr | — | — | — | $2.40 | $1.67 | — | — | | Salad | — | — | — | — | — | $0.33 | $0.50 | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded, rounded to the cent. Regions: AWS us-east-1, Azure East US, Google Cloud us-central1 (Iowa). Salad's price is its High Priority tier on 29 October 2026. ## How has the GPU rental price changed over time? This index cannot say yet. October 2026 is its first edition, and it does not borrow history from other sources, because their rules differ. The November 2026 edition will show the first month-on-month change, with the same method. One change is already announced. Salad raises its High Priority price on 1 November 2026: the RTX 4090 from $0.33 to $0.44 per hour, and the RTX 5090 from $0.50 to $0.60. Both were the lowest on-demand prices for those GPUs in this edition. ## How do AWS, Google Cloud and Azure change the median? They pull it up. Their GPU prices include the whole VM (vCPUs, RAM and local SSD), and their H100, H200 and B200 machines are mostly 8-GPU VMs. Without the three hyperscalers, the medians on 29 October 2026 were: - **H100 80GB**: $3.49 across 13 providers, against $4.16 with them. - **H200 141GB**: $4.80 across 11, against $5.29. - **B200**: $8.25 across 8, against $8.50. - **A100 80GB**: $2.00 across 11, against $2.23. - **L40S**: $1.62 across 10, against $1.67. The index keeps them in: they are self-serve on-demand list prices like the others. ## How is the index calculated? The same rule for every GPU and every provider: 1. **Sources.** Each price was read on 29 October 2026 on the provider's own pricing page, or on the public price feed or API that page loads. No aggregator is used. 2. **One price per provider.** The provider's lowest self-serve on-demand price for that GPU, across all its configurations. The H100 groups SXM, PCIe, NVL and unstated form factors; the H200 groups SXM and NVL. 3. **Per GPU.** A price listed only for a multi-GPU instance is divided by its GPU count: CoreWeave, Google Cloud, AWS and Azure 8-GPU machines, OVHcloud's 8× H200 and Massed Compute's 8× B200. 4. **Hyperscalers** keep the whole VM in the price, as they sell it. 5. **Statistics.** The median of an even count is the mean of the two middle prices. The 25th percentile uses linear interpolation between ranks, the method of a spreadsheet's PERCENTILE.INC. 6. **Exclusions.** Spot, preemptible, community and marketplace prices; contact-sales prices; stale pages (TensorDock's page dated July 2024, Spheron's RTX 4090 marked "Last known Sep 17"); Voltage Park's "starting at $1.99" H100, because its pricing cards say "Contact for pricing". 7. **Vast.ai is excluded.** Its public price feed allows display with attribution, but its licence asks for a data licence before use in an index or benchmark. Fluidstack, Genesis Cloud, io.net and Prime Intellect had no readable public price list on 29 October 2026, and Hetzner listed no GPU from the index. Paperspace and DigitalOcean count as one provider. ## Where do GPUwithCrypto prices sit against the index? GPUwithCrypto is not in the index: its prices are set from the same price check. Each GPU is priced at least 15% under a named reference: the on-demand price of the same GPU at a self-serve provider, read on 29 October 2026 (H100: Massed Compute at $2.73 per GPU-hour; each GPU's reference is named on its price page). Spot, interruptible, marketplace and contact-sales prices are not used as references, and some of them are lower. We do not claim to be the cheapest for every GPU. | GPU | Index lowest | Index median | GPUwithCrypto | | --- | --- | --- | --- | | H100 80GB | $2.73 | $4.16 | **$2.32/hour** | | H200 141GB | $3.62 | $5.29 | **$3.07/hour** | | B200 | $5.43 | $8.50 | **$4.61/hour** | | A100 80GB | $1.09 | $2.23 | **$0.92/hour** | | L40S | $0.97 | $1.67 | **$0.82/hour** | | RTX 4090 | $0.33 | $0.61 | **$0.28/hour** | | RTX 5090 | $0.50 | $0.84 | **$0.42/hour** | *Prices checked on October 29, 2026.* GPUwithCrypto prices are on-demand, billed per minute, paid in BTC, USDT, USDC, ETH, SOL, LTC, XMR with no KYC. Full offer on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## When is the next edition? The index is planned monthly. Each edition keeps its own check date, so a figure quoted from this page stays tied to 29 October 2026. To quote it, give the date and link to this page. Per-GPU detail, with spot and marketplace prices listed apart, is on the [GPU price pages](https://gpuwithcrypto.com/gpu-prices/ "How much does it cost to rent a GPU per hour, day or month?") and the [cheapest GPU cloud page](https://gpuwithcrypto.com/cheapest-gpu-cloud/ "What is the cheapest GPU cloud for AI right now?"). ## GPU price index: questions ### What is the average price to rent an H100 per hour? On 29 October 2026 the median on-demand H100 80GB cost $4.16 per GPU-hour across 16 providers. The middle half of providers charged between $3.20 and $5.93; without AWS, Google Cloud and Azure the median was $3.49. ### Is there a free GPU rental price index? Yes, this page. It is free to read and to quote with the date and a link. Commercial indexes also exist, but this one publishes its full provider list and method. ### Why is Vast.ai not in the index? Vast.ai's public price feed licence allows display with attribution, but asks for a data licence before its prices are used in an index or benchmark. Its marketplace offers are also not comparable with fixed on-demand list prices. ### Why does the index leave out spot prices? Spot and preemptible GPUs can be taken back, and marketplace prices change with each host, so they do not compare with an on-demand list price. They are listed separately on the [cheapest GPU cloud page](https://gpuwithcrypto.com/cheapest-gpu-cloud/ "What is the cheapest GPU cloud for AI right now?"). ### How often is the GPU price index updated? Monthly. October 2026 is the first edition; the November 2026 edition will add the first month-on-month change. ## Sources 1. [Pricing (VM)](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 3. [Public Cloud prices](https://www.ovhcloud.com/en/public-cloud/prices/) — OVHcloud (read 2026-10-29) 4. [GPU marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29) 5. [Pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 6. [Pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 7. [Pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 8. [Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29) 9. [GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 10. [Prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 11. [Pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 12. [Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 13. [Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 14. [EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — AWS (read 2026-10-29) 15. [Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 16. [Azure Retail Prices API, East US](https://prices.azure.com/api/retail/prices?$filter=armRegionName%20eq%20'eastus'%20and%20serviceName%20eq%20'Virtual%20Machines') — Microsoft (read 2026-10-29) 17. [Plans API](https://api.vultr.com/v2/plans?type=all&per_page=500) — Vultr (read 2026-10-29) 18. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 19. [Pricing (Contact for pricing; excluded)](https://www.voltagepark.com/pricing) — Voltage Park (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does it cost to rent a GPU per hour, day or month? > How much does it cost to rent a GPU? On-demand prices from $0.33 to $14.24 an hour across 18 providers on 29 Oct 2026, and ours in crypto with no KYC. - URL: https://gpuwithcrypto.com/gpu-prices/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) On-demand prices of seven NVIDIA GPUs across 18 providers, the cost of a day and a month, and how spot, marketplace and crypto change the picture. **Quick answer — How much does it cost to rent a GPU per hour, day or month?** As of 29 October 2026, renting a cloud GPU on demand costs from $0.33 per hour (RTX 4090) to $14.24 (B200, AWS 8-GPU instance). The median was $2.23 for an A100 80GB and $4.16 for an H100. On **GPUwithCrypto**, paid in crypto with no KYC, an H100 costs $2.32/hour: $55.68 a day or $1,670.40 for 720 hours. ## How much does each GPU cost per hour, day and month? Lowest and median self-serve on-demand price per GPU-hour on 29 October 2026. Next to them: our price for one hour, 24 hours and 720 hours, each GPU priced at least 15% under a named on-demand reference. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Click a GPU for every provider row. | GPU | Lowest on-demand (provider) | Median on-demand | Providers | GPUwithCrypto per hour | GPUwithCrypto 24 h | GPUwithCrypto 720 h | Median price × 720 h | | --- | --- | --- | --- | --- | --- | --- | --- | | [H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | $2.73 (Massed Compute) | $4.16 | 16 | **$2.32/hour** | $55.68 | $1,670.40 | $2,991.60 | | [H200 141GB](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") | $3.62 (Massed Compute) | $5.29 | 13 | **$3.07/hour** | $73.68 | $2,210.40 | $3,808.80 | | [B200 180GB](https://gpuwithcrypto.com/gpu-prices/b200/ "How much does a B200 cost per hour?") | $5.43 (Massed Compute) | $8.50 | 9 | **$4.61/hour** | $110.64 | $3,319.20 | $6,120.00 | | [A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | $1.09 (Thunder Compute) | $2.23 | 13 | **$0.92/hour** | $22.08 | $662.40 | $1,602.72 | | [L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | $0.97 (Massed Compute) | $1.67 | 11 | **$0.82/hour** | $19.68 | $590.40 | $1,203.12 | | [RTX 4090 24GB](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") | $0.33 (Salad) | $0.61 | 2 | **$0.28/hour** | $6.72 | $201.60 | $439.20 | | [RTX 5090 32GB](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") | $0.50 (Salad) | $0.84 | 3 | **$0.42/hour** | $10.08 | $302.40 | $604.80 | *Prices checked on October 29, 2026.* USD per GPU, taxes excluded. Each provider counts once, with its lowest on-demand price for that GPU in any form factor; multi-GPU instance prices are divided by the GPU count. Spot, marketplace, reserved and contact-sales prices are not included. Salad's RTX prices rise on 1 November 2026. ## Which GPU page do you need? - **[H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?")** — Training and fine-tuning up to 70B parameters, fast LLM inference. GPUwithCrypto: $2.32/hour. - **[H200 141GB](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?")** — Large-model inference and long contexts: a 70B model in FP16 fits on two GPUs. GPUwithCrypto: $3.07/hour. - **[B200 180GB](https://gpuwithcrypto.com/gpu-prices/b200/ "How much does a B200 cost per hour?")** — Frontier-scale training and FP4/FP8 inference. GPUwithCrypto: $4.61/hour. - **[A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?")** — Fine-tuning and inference on a budget, the most rented data-center GPU. GPUwithCrypto: $0.92/hour. - **[L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?")** — Image and video generation, inference of models up to about 30B parameters. GPUwithCrypto: $0.82/hour. - **[RTX 4090 24GB](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?")** — ComfyUI, Stable Diffusion, Flux and LoRA training at the lowest hourly price. GPUwithCrypto: $0.28/hour. - **[RTX 5090 32GB](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?")** — Image and video generation and small-model inference with 32 GB of VRAM. GPUwithCrypto: $0.42/hour. ## What is the difference between on-demand, spot and marketplace prices? On-demand is a fixed list price for a GPU nobody can take back. Spot and marketplace prices are lower because the GPU can be interrupted, or because an independent host sets the price. The four kinds of price we saw on 29 October 2026: - **On-demand:** a fixed hourly rate; the instance runs until you stop it. Example: Massed Compute's H100 80GB at $2.73. - **Spot or preemptible:** the provider can stop the instance when it needs the GPU back. Nebius lists preemptible H100s from $0.79 per GPU-hour. Google Cloud's spot H100 cost $6.59 per GPU, against $11.06 on demand. - **Marketplace or community:** independent hosts rent out their machines and set the price. Vast.ai's public price feed showed a rentable RTX 4090 at $0.13 and an H100 SXM at $1.73. Akash's lowest recent provider bid for an H100 SXM was $2.04. - **Reserved:** a lower rate for a commitment. DigitalOcean's 12-month reserved H100 costs $3.26 per GPU-hour. Our prices are compared with on-demand prices only. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. The [spot GPU guide](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?") covers when an interruptible GPU is worth it. ## Why does the same GPU cost up to five times more at some providers? Form factor, instance size, region and what the price includes explain most of the gap. On 29 October 2026, the most expensive self-serve on-demand H100 cost 4.5 times the cheapest. The gap was 4.7 times for the A100 80GB, 3.2 times for the L40S and 2.9 times for the H200. - **Form factor:** RunPod listed the H100 PCIe at $2.89, the H100 NVL at $3.19 and the H100 SXM at $3.99. - **Instance size:** Lambda charged $3.99 per GPU in an 8-GPU H100 SXM instance and $4.29 for a single GPU. - **Whole VM:** hyperscaler prices include vCPUs, RAM and local SSD. Google Cloud's 8-GPU H100 VM works out at $11.06 per GPU. - **Region:** the hyperscaler rows are AWS US East, Azure East US and Google Cloud Iowa; other regions differ. Each GPU page lists the configuration behind every price, so you compare like with like. ## What does a GB of GPU memory cost per hour? GPUwithCrypto's price check of 29 October 2026 shows a 2× gap in the cost of GPU memory. At median prices, a GB of memory costs $0.052 per hour on an H100 and $0.025 on an RTX 4090. The table divides each median on-demand price by the GPU's memory. | GPU | Memory | Median on-demand per hour | Median per GB of memory per hour | | --- | --- | --- | --- | | H100 | 80 GB | $4.16 | $0.052 | | B200 | 180 GB | $8.50 | $0.047 | | H200 | 141 GB | $5.29 | $0.038 | | L40S | 48 GB | $1.67 | $0.035 | | A100 80GB | 80 GB | $2.23 | $0.028 | | RTX 5090 | 32 GB | $0.84 | $0.026 | | RTX 4090 | 24 GB | $0.61 | $0.025 | *Prices checked on October 29, 2026.* Memory sizes from NVIDIA's product pages; the B200 figure is NVIDIA's 1,440 GB per 8-GPU board divided by 8. Memory is not the only cost driver: bandwidth and FP8 support differ too. ## Which is the cheapest GPU to rent per hour? The RTX 4090 is the cheapest GPU in our check. Salad's High Priority tier cost $0.33 per hour on 29 October 2026 and rises to $0.44 on 1 November. On GPUwithCrypto it costs $0.28/hour, with 24 GB of memory. The cheapest data-center GPU is the A100 80GB: $1.09 at Thunder Compute, and $0.92/hour here. It has no FP8 Tensor Cores; the L40S, H100 and H200 do. The right GPU is the cheapest one your model fits on. By EleutherAI's 1.2× rule, a 70B model needs about 168 GB in BF16, 84 GB in FP8 and 42 GB in 4-bit. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the full table, and the [cheapest GPU cloud ranking](https://gpuwithcrypto.com/cheapest-gpu-cloud/ "What is the cheapest GPU cloud for AI right now?") compares providers. ## How do I work out the cost of a day or a month? Multiply the hourly price by 24 for a day and by 720 for a 30-day month. At the median on-demand price of 29 October 2026, an H100 cost $99.72 a day and $2,991.60 a month. On GPUwithCrypto the same GPU costs $55.68 a day and $1,670.40 for 720 hours. A job rarely runs exactly 24 hours, so the billing step matters. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. Every block is on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## Which GPU clouds accept crypto, and which ask for ID? Of the 33 GPU clouds we checked on 29 October 2026, 14 accept crypto, and 4 ask for ID in some cases. None of the 13 cited GPU price and comparison pages we reviewed that day showed this next to the price, so here it is: - **Crypto, with KYC before the first crypto payment:** RunPod. - **Crypto, no ID step documented:** Vast.ai (BitPay, Crypto.com), Massed Compute (BTC), Salad (USDC, USDT and RENDER on Solana), Vultr (BitPay), Spheron (USDC) and DigitalOcean (stablecoins, limited availability). - **No crypto:** Lambda, Nebius, Verda, Jarvislabs, Thunder Compute, Crusoe, OVHcloud, AWS, Google Cloud and Azure, among others. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR and asks for no ID, no selfie and no card. The [crypto GPU cloud comparison](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") lists coins, processors and refund rules, and the [no-KYC GPU cloud guide](https://gpuwithcrypto.com/no-kyc-gpu-cloud/ "Is there a GPU cloud with no KYC?") explains who asks for ID and when. ## Questions about GPU rental costs ### How much does it cost to rent a GPU for a day? From $6.72 for an RTX 4090 to $55.68 for an H100 on GPUwithCrypto, for 24 hours of one GPU. At the median on-demand price of 29 October 2026, an H100 day cost $99.72 elsewhere. ### How much does it cost to rent a GPU for a month? A 30-day month is 720 GPU-hours. On GPUwithCrypto that is $662.40 for an A100 80GB and $1,670.40 for an H100. At the median on-demand price of 29 October 2026, an H100 month cost $2,991.60. ### Is renting a GPU by the minute cheaper than by the hour? Yes for short or uneven jobs, because you stop paying when the job ends. A 90-minute run billed per hour costs two hours. Per-second billing (RunPod, Salad, Vast.ai) and per-minute billing (Lambda, Jarvislabs, GPUwithCrypto) avoid that. ### Which GPU do I need for a 70B model? In BF16 a 70B model needs about 168 GB: three 80 GB GPUs or two H200s. In FP8 it needs about 84 GB and fits on one H200. In 4-bit it needs about 42 GB and fits on one 48 GB L40S. ### Do GPU rental prices include CPU, RAM and storage? It depends on the provider. Hyperscaler prices cover the whole VM, Thunder Compute includes 4 to 8 vCPUs, and RunPod bills container disk at $0.10 per GB per month. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. ## Sources 1. [Massed Compute — VM pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Thunder Compute — Pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 3. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 4. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 5. [Lambda — Pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 6. [Nebius — Prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 7. [Google Cloud — Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 8. [AWS — EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — Amazon Web Services (read 2026-10-29) 9. [DigitalOcean — GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 10. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 11. [Akash Network — GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) 12. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 13. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) 14. [NVIDIA DGX B200](https://www.nvidia.com/en-us/data-center/dgx-b200/) — NVIDIA (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does an A100 cost per hour? > How much does an A100 cost per hour? 80GB from $1.09, 40GB from $0.89 on demand on 29 Oct 2026. Our 80GB: at least 15% under $1.09, in crypto, no KYC. - URL: https://gpuwithcrypto.com/gpu-prices/a100/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Every on-demand A100 80GB price we found, the 40 GB vs 80 GB gap, spot and marketplace prices, and which providers take crypto. No ID and no card: top up in crypto, start an A100 as soon as the top-up is credited. **Quick answer — How much does an A100 cost per hour?** As of 29 October 2026, an NVIDIA A100 80GB costs from $1.09 to $5.07 per GPU-hour on demand across 13 providers, with a median of $2.23. The A100 40GB starts at $0.89, with a median of $1.99. On **GPUwithCrypto** an A100 80GB costs $0.92/hour, billed per minute and paid in crypto with no KYC. ## How much does an A100 80GB cost per hour at each provider? One row per provider at its cheapest on-demand A100 80GB configuration, sorted by price. The most expensive on-demand A100 80GB we found (Google Cloud, $5.07) costs 4.7 times the cheapest ($1.09 at Thunder Compute). | Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC | | --- | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** | A100 80GB, 1 to 8 GPUs; SXM or PCIe shown per host | **$0.92** | — (on-demand only) | Per minute | Yes | No | | Thunder Compute | A100 80GB (form factor not stated), 1 GPU | $1.09 | — | Per minute | No | None documented | | Massed Compute | A100 80GB (form factor not stated), 1 GPU (SXM: $1.38) | $1.35 | — | Hourly rate (increment not stated) | Yes (BTC) | None documented | | Jarvislabs | A100 80GB (form factor not stated), 1 GPU | $1.49 | $0.89 (spot) | Per minute | No | None documented | | Spheron | A100 80GB (form factor not stated) | $1.69 | — | Per minute, 20-minute minimum | Yes (USDC) | None documented | | RunPod | A100 80GB PCIe or SXM, 1 GPU | $1.79 | $1.19 PCIe, $1.39 SXM (Community Cloud) | Per second | Yes | KYC before first crypto payment | | Crusoe | A100 80GB PCIe (SXM: $2.30) | $2.00 | Contact sales | Per hour or per second | No | None documented | | Verda (formerly DataCrunch) | A100 80GB SXM, 1 GPU | $2.23 | $1.11 (spot) | — | No | None documented | | Vultr | A100 80GB (form factor not stated), 1 GPU | $2.40 | — | Hourly | Yes (BitPay) | None documented | | CoreWeave | A100 80GB, 8-GPU instance ÷ 8 | $2.70 | $1.21 (spot) | — | Unknown | Unknown | | Lambda | A100 80GB SXM, 8 GPUs | $2.79 | — | Per minute | No (cards only) | None documented | | OVHcloud | A100 80GB (form factor not stated), 1 GPU | $3.07 | — | Per second, 1-minute minimum | No | Sometimes | | AWS | A100 80GB, 8 GPUs (p4de.24xlarge) ÷ 8 | $3.43 | — | — | No | None documented | | Google Cloud | A100 80GB, 1 GPU (a2-ultragpu-1g) | $5.07 | $3.01 (spot) | Per second, 1-minute minimum | No | None documented | | Vast.ai | A100 80GB PCIe and SXM, marketplace | — | $0.34 PCIe, $0.48 SXM (Vast.ai's public price feed, cheapest rentable offers) | Per second | Yes (BitPay, Crypto.com) | None documented | | Akash Network | A100 80GB SXM, decentralized marketplace | — | $1.07 (lowest recent provider bid) | — | Yes (AKT, ACT) | None documented | | Shadeform | A100 80GB SXM, 1 GPU, resold from other clouds | — | $1.50 (cheapest available listing) | — | Yes (USDC, per Stripe) | None documented | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded, rounded to the cent. Hyperscaler prices cover the whole VM (AWS US East, Google Cloud Iowa). Vultr's figure comes from its public plans API, which did not confirm availability. Paperspace showed only a 3-year commitment price for its A100 80GB, so it is not listed. 'None documented' means no ID step in the official pages we read. ## A100 40GB vs 80GB: how much more does the 80GB cost? According to GPUwithCrypto's price check of 29 October 2026, the same provider charges 25% to 67% more for the 80GB A100 than the 40GB. The medians differ by only 12%, because the cheapest clouds list only the 80 GB card. | Provider | A100 40GB | A100 80GB | 80GB premium | | --- | --- | --- | --- | | Jarvislabs | $0.89 | $1.49 | +67% | | Verda (formerly DataCrunch) | $1.59 (SXM) | $2.23 (SXM) | +40% | | Lambda | $1.99 (SXM, 8 GPUs) | $2.79 (SXM, 8 GPUs) | +40% | | Google Cloud | $3.67 | $5.07 | +38% | | AWS | $2.74 (p4d, 8 GPUs ÷ 8) | $3.43 (p4de, 8 GPUs ÷ 8) | +25% | | Median of all providers | $1.99 (5 providers) | $2.23 (13 providers) | +12% | *Prices checked on October 29, 2026.* The 80 GB card also has faster memory: 2,039 GB/s on the SXM version against 1,555 GB/s for the 40 GB SXM4. Compute is the same. ## Where is an A100 cheaper than ours, and why? On marketplaces, on spot, and in the 40 GB version. We say so because the cheapest option depends on your job. - **Marketplace:** the cheapest rentable A100 80GB PCIe in Vast.ai's public price feed cost $0.34 per hour, among 91 rentable offers. Independent hosts set these prices and run the machines. - **Spot:** Jarvislabs sells interruptible A100 80GB hours at $0.89. The provider can stop them. - **40 GB:** Jarvislabs' on-demand A100 40GB costs $0.89. If your model fits in 40 GB, it is cheaper than our 80 GB card. For an 80 GB A100 that nobody can take back, our $0.92/hour is 59% below the median of 13 providers. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. The reference is Thunder Compute's A100 80GB at $1.09, whose page does not state the form factor. ## How much does an A100 cost per day, week or month? Our A100 80GB price next to the lowest and the median self-serve on-demand price of 29 October 2026, for the same hours. | Period | GPU-hours | GPUwithCrypto | Lowest elsewhere (Thunder Compute) | Median elsewhere | | --- | --- | --- | --- | --- | | 1 hour | 1 | **$0.92** | $1.09 | $2.23 | | 1 day | 24 | **$22.08** | $26.16 | $53.42 | | 1 week | 168 | **$154.56** | $183.12 | $373.97 | | 30 days | 720 | **$662.40** | $784.80 | $1,602.72 | *Prices checked on October 29, 2026.* The median is $2.226 per GPU-hour before rounding. Our day, week and month match the 24, 168 and 720 GPU-hour blocks. ## What are the A100's specs? NVIDIA's official figures. NVIDIA prints both dense and sparse Tensor values for the A100; compare the dense ones with other GPUs. | Spec | A100 80GB SXM | A100 80GB PCIe | A100 40GB SXM4 | | --- | --- | --- | --- | | Memory | 80 GB HBM2e | 80 GB HBM2e | 40 GB HBM2 | | Memory bandwidth | 2,039 GB/s | 1,935 GB/s | 1,555 GB/s | | FP16 / BF16 Tensor, dense | 312 TFLOPS (624 sparse) | 312 TFLOPS (624 sparse) | 312 TFLOPS (624 sparse) | | TF32 Tensor, dense | 156 TFLOPS (312 sparse) | 156 TFLOPS (312 sparse) | 156 TFLOPS (312 sparse) | | FP8 Tensor | Not supported | Not supported | Not supported | | GPU-to-GPU link | NVLink 600 GB/s | NVLink bridge 600 GB/s | NVLink 600 GB/s | | Max power | 400 W | 300 W | 400 W | ## What fits on one A100 40GB or 80GB? Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb). The A100 has no FP8 Tensor Cores; the L40S, H100 and H200 do. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the formulas. | Model and job | Memory needed | A100 40GB | A100 80GB | | --- | --- | --- | --- | | Llama 3.1 8B inference, BF16 | ≈ 19.2 GB | Yes | Yes | | Llama 3.1 8B LoRA fine-tune | ≈ 19.2 GB | Yes | Yes | | Mistral 7B inference, BF16 | ≈ 17.5 GB | Yes | Yes | | FLUX.1 [dev] transformer, BF16 weights | 24 GB, plus text encoders and VAE | Yes | Yes | | Llama 3.1 70B inference, 4-bit | ≈ 42 GB | No | Yes | | Qwen2.5 72B inference, 4-bit | ≈ 43.6 GB | No | Yes | | Llama 3.1 70B QLoRA fine-tune | ≈ 52.5 GB | No | Yes | | Llama 3.1 8B full fine-tune | ≈ 128 to 144 GB before activations | No | No | | Llama 3.1 70B inference, BF16 | ≈ 168 GB | No | No | ## What does an A100 80GB cost on GPUwithCrypto? $0.92/hour, at least 15% under its named on-demand reference: Thunder Compute at $1.09, read on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $9.20 | $0.92/hour | | Day | 24 GPU-hours | $22.08 | $0.92/hour | | Week | 168 GPU-hours | $154.56 | $0.92/hour | | Month | 720 GPU-hours | $662.40 | $0.92/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. - Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. - Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min · XMR ≈ 20 min ## A100 price questions ### What is the cheapest A100 rental? On 29 October 2026 the lowest self-serve on-demand A100 80GB price we found was $1.09 at Thunder Compute, and GPUwithCrypto charges $0.92. Jarvislabs' A100 40GB cost $0.89. Marketplace offers in Vast.ai's public price feed started at $0.34, but hosts set and can change those prices. ### A100 40GB vs 80GB: which should I rent? Rent the 80 GB card for 70B models in 4-bit or QLoRA, which need 42 to 53 GB. The 40 GB card is enough for 7B and 8B models in BF16. The same provider charged 25% to 67% more for the 80 GB version on 29 October 2026. ### How much does an A100 cost per month? 720 GPU-hours of an A100 80GB cost $662.40 on GPUwithCrypto. At the median on-demand price of 29 October 2026 the same month cost $1,602.72, and $784.80 at the lowest. ### Is the A100 still worth renting? Yes for BF16 training and inference that fits in 80 GB: it was the cheapest 80 GB data-center GPU in our 29 October 2026 check. For FP8 models pick an L40S, H100 or H200. See [L40S vs A100](https://gpuwithcrypto.com/compare/l40s-vs-a100/ "L40S vs A100 for inference and fine-tuning: which should I rent?") and [H100 vs A100](https://gpuwithcrypto.com/compare/h100-vs-a100/ "H100 vs A100: which should I rent?"). ### Can I rent an A100 with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID, no selfie and no card. Of the 13 self-serve A100 80GB providers we checked, 4 accept crypto: Massed Compute, Spheron, Vultr and RunPod. RunPod asks for KYC before the first crypto payment. ## Sources 1. [Thunder Compute — Pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 2. [Massed Compute — VM pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 3. [Jarvislabs — Pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 4. [Spheron — Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 5. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 6. [Crusoe — Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29) 7. [Verda — Pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 8. [Vultr — Public plans API](https://api.vultr.com/v2/plans?type=all&per_page=500) — Vultr (read 2026-10-29) 9. [CoreWeave — Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 10. [Lambda — Pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 11. [OVHcloud — Public Cloud prices](https://www.ovhcloud.com/en/public-cloud/prices/) — OVHcloud (read 2026-10-29) 12. [AWS — EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — Amazon Web Services (read 2026-10-29) 13. [Google Cloud — Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 14. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 15. [Akash Network — GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) 16. [Shadeform — Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29) 17. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 18. [NVIDIA A100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/a100/) — NVIDIA (read 2026-10-29) 19. [NVIDIA Ampere architecture whitepaper (A100 40GB)](https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/nvidia-ampere-architecture-whitepaper.pdf) — NVIDIA (read 2026-10-29) 20. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does a B200 cost per hour? > How much does a B200 cost per hour? $5.43 to $14.24 on demand at 9 providers on 29 Oct 2026, and $4.61/hour here in crypto, no KYC. - URL: https://gpuwithcrypto.com/gpu-prices/b200/ - Last updated: 2026-11-03 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Every on-demand B200 price we found, the spot prices that beat ours, what fits in 180 GB, and who takes crypto or asks for ID. No ID and no card: top up in crypto, start a B200 as soon as the top-up is credited. **Quick answer — How much does a B200 cost per hour?** As of 29 October 2026, an NVIDIA B200 with 180 GB HBM3e costs from $5.43 to $14.24 per GPU-hour on demand across 9 providers, with a median of $8.50. On **GPUwithCrypto** it costs $4.61/hour, billed per minute and paid in crypto with no KYC. Interruptible B200s go lower: Nebius lists preemptible B200s from $0.99, and CoreWeave and Verda sell spot B200s at $4.26 and $4.39. ## How much does a B200 cost per hour at each provider? One row per provider at its cheapest on-demand B200 configuration, sorted by price. The most expensive on-demand B200 we found (AWS, $14.24 per GPU in an 8-GPU instance) costs 2.6 times the cheapest. | Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC | | --- | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** | B200 180GB, 1 to 8 GPUs; GPU count shown per host | **$4.61** | — (on-demand only) | Per minute | Yes | No | | Massed Compute | B200 SXM6, 8-GPU node ($43.46/hour) ÷ 8 | $5.43 | — | Hourly rate (increment not stated) | Yes (BTC) | None documented | | Hyperbolic | B200 (form factor and GPU count not stated) | $5.99 | — | — | Unknown | None documented | | Lambda | B200 SXM6, 8-GPU instance (1 GPU: $6.99; 2: $6.89; 4: $6.79) | $6.69 | — | Per minute | No (cards only) | None documented | | RunPod | B200 (form factor not stated), 1 GPU | $7.99 | $5.98 (Community Cloud) | Per second | Yes | KYC before first crypto payment | | Nebius | B200 SXM (HGX) | $8.50 | From $0.99 (preemptible) | — | No | None documented | | CoreWeave | B200 SXM (HGX), 8-GPU instance ÷ 8 | $8.60 | $4.26 (spot) | — | Unknown | Unknown | | Verda (formerly DataCrunch) | B200 SXM6, 1 GPU | $8.78 | $4.39 (spot) | — | No | None documented | | Spheron | B200 (form factor not stated) | $11.12 | $5.95 (spot) | Per minute, 20-minute minimum | Yes (USDC) | None documented | | AWS | B200, 8 GPUs (p6-b200.48xlarge) ÷ 8 | $14.24 | — | — | No | None documented | | Google Cloud | B200, 8-GPU VM ÷ 8 (a4-highgpu-8g) | Shown as N/A | $4.95 (spot) | Per second, 1-minute minimum | No | None documented | | Crusoe | B200 180GB (HGX) | Contact sales | — | Per hour or per second | No | None documented | | Vast.ai | B200, marketplace | — | $7.50 (Vast.ai's public price feed, cheapest of 50 rentable offers) | Per second | Yes (BitPay, Crypto.com) | None documented | | Shadeform | B200, 8 GPUs, resold from other clouds | — | None available (11 listings, lowest listed $3.74) | — | Yes (USDC, per Stripe) | None documented | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded, rounded to the cent. Multi-GPU prices are divided by the GPU count. Hyperscaler prices cover the whole VM (AWS US East, Google Cloud Iowa). 'None documented' means no ID step in the official sign-up, billing or terms pages we read, not a promise that a provider never asks. RunPod's Community Cloud is community-hosted. Google Cloud, Crusoe, Vast.ai and Shadeform are not in the median. ## Is the B200 worth the extra cost over an H200? Yes when your model needs more than 141 GB per GPU, or when your job runs in FP8 or FP4. According to GPUwithCrypto's price check of 29 October 2026, a median on-demand B200 costs 61% more per hour than a median H200 ($8.50 against $5.29). For that it has 28% more memory (180 GB against 141 GB), 67% more bandwidth (8,000 against 4,800 GB/s) and 2.3 times the dense FP8 rating (4,500 against 1,979 TFLOPS). On paper, B200 compute is cheaper: at median prices, a dense FP8 TFLOPS costs 29% less per hour on a B200 than on an H200. These are NVIDIA's peak ratings, not measured speeds. The B200 also has FP4 Tensor Cores, rated 9,000 TFLOPS dense; NVIDIA gives the H200 no FP4 rating. Memory is the clearest gain. Llama 3.1 70B in BF16 needs about 168 GB: one B200, or two H200s. The B200 is the only GPU we rent that holds it on one card. When your model already fits in 141 GB, the H200 costs less per hour: $3.07/hour here. The [H200 price page](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") has every H200 row, and the [H100 vs H200 comparison](https://gpuwithcrypto.com/compare/h100-vs-h200/ "H100 vs H200: is the H200 worth the extra cost?") covers the step below. ## Why are some spot and preemptible B200s cheaper than ours? Because the provider can take them back. On 29 October 2026, three interruptible B200 prices were below our $4.61: - **Preemptible:** Nebius listed B200s from $0.99 per GPU-hour. The rate floats, capped one cent below its on-demand price, and Nebius can reclaim the GPU. - **Spot:** CoreWeave's 8-GPU B200 instance cost $34.11 per hour on spot, or $4.26 per GPU. Verda's single-GPU spot B200 cost $4.39. - **Above ours:** Google Cloud's spot B200 cost $4.95 per GPU, Spheron's spot $5.95 and RunPod's Community Cloud $5.98. The cheapest rentable B200 in Vast.ai's public price feed was $7.50, among 50 offers with a median of $8.75. Shadeform's API listed B200s from $3.74 per GPU-hour, but none of its 11 on-demand listings was available when we read it. Spot suits jobs that save checkpoints and can restart. A long training run or a live inference server needs a GPU that stays put. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. The [spot GPU guide](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?") explains when an interruption is worth the saving. ## Which B200 sets the reference price, and can you rent a single GPU? The reference is Massed Compute's 8-GPU B200 SXM6 node at $43.46 per hour, or $5.43 per GPU-hour. Massed Compute lists no smaller B200 machine, and the AWS price is for an 8-GPU instance. Single B200s exist, at a higher rate per GPU. Lambda charges $6.99 for one GPU and $6.69 per GPU in an 8-GPU instance. RunPod lists one B200 at $7.99 and Verda at $8.78. Three clouds show no self-serve on-demand B200 price. Crusoe says 'Contact sales', Google Cloud lists the on-demand price of its 8-GPU B200 VM as N/A, and Vultr's API marks its 8-GPU B200 plan as not deployable on demand. None of them counts in the median. GPUwithCrypto has one B200 price for every host. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## How much does a B200 cost per day, week or month? Our price for one GPU next to the lowest and the median self-serve on-demand price of 29 October 2026, for the same number of hours. Our hourly rate is 46% under the median. | Period | GPU-hours | GPUwithCrypto | Lowest elsewhere (Massed Compute) | Median elsewhere | | --- | --- | --- | --- | --- | | 1 hour | 1 | **$4.61** | $5.43 | $8.50 | | 1 day | 24 | **$110.64** | $130.38 | $204.00 | | 1 week | 168 | **$774.48** | $912.66 | $1,428.00 | | 30 days | 720 | **$3,319.20** | $3,911.40 | $6,120.00 | *Prices checked on October 29, 2026.* Massed Compute's rate is $5.4325 per GPU-hour before rounding ($43.46 ÷ 8). Our day, week and month match the 24, 168 and 720 GPU-hour blocks. ## What are the B200's specs? NVIDIA publishes B200 figures only for its 8-GPU HGX B200 and DGX B200 systems: 1,440 GB of memory, 64 TB/s and 72 PFLOPS of sparse FP8. The B200 column divides those totals by 8; dense is half the sparse figure, per NVIDIA's own footnote. The H200 SXM is shown for comparison. | Spec | B200 (per GPU, HGX B200) | H200 SXM | | --- | --- | --- | | Architecture, launch | Blackwell, 2024 | Hopper, 2023 | | Memory | 180 GB HBM3e | 141 GB HBM3e | | Memory bandwidth | 8,000 GB/s | 4,800 GB/s | | FP4 Tensor, dense | 9,000 TFLOPS (18,000 sparse) | — | | FP8 Tensor, dense | 4,500 TFLOPS (9,000 sparse) | 1,979 TFLOPS (3,958 sparse) | | FP16 / BF16 Tensor, dense | 2,250 TFLOPS (4,500 sparse) | 989.5 TFLOPS (1,979 sparse) | | TF32 Tensor, dense | 1,125 TFLOPS (2,250 sparse) | 494.5 TFLOPS (989 sparse) | | FP32 | 75 TFLOPS | 67 TFLOPS | | GPU-to-GPU link | NVLink 1,800 GB/s | NVLink 900 GB/s | | Max power | 1,000 W | 700 W | NVIDIA states the B200's NVLink speed per GPU; the other B200 figures are 8-GPU totals ÷ 8. The B200 power figure is the per-GPU maximum that nvidia-smi reports in a DGX B200. ## What fits on one B200? Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb) and weights plus gradients and optimizer for fine-tuning. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the formulas and more models. | Model and job | Memory needed | One B200 180GB? | | --- | --- | --- | | Llama 3.1 70B inference, BF16 | ≈ 168 GB | Yes | | Llama 3.1 70B, FP8 weights plus two full 128K-token contexts (BF16 cache) | ≈ 156 GB | Yes | | Qwen2.5 72B inference, FP8 | ≈ 87.2 GB | Yes | | Llama 3.1 70B LoRA fine-tune | ≈ 140 to 168 GB | Yes | | Llama 3.1 8B full fine-tune | ≈ 128 to 144 GB before activations | Yes, with 36 to 52 GB left for activations | | Llama 3.1 405B inference, 4-bit | ≈ 243 GB | No: 2× B200 | | Llama 3.1 405B inference, FP8 | ≈ 486 GB | No: 3× B200 | | DeepSeek-V3 or R1 inference, FP8 | ≈ 805 GB | No: one 8× B200 node (1,440 GB) | Two 128K-token contexts for Llama 3.1 70B add 2 × 42.9 GB of BF16 KV cache to its 70 GB of FP8 weights. Only 37B of DeepSeek-V3's 671B parameters are active per token, but all of them must sit in memory. ## What does a B200 cost on GPUwithCrypto? $4.61/hour, at least 15% under a named on-demand reference read on 29 October 2026: Massed Compute at $5.43. We do not claim to be the cheapest for every GPU; spot and marketplace offers can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $46.10 | $4.61/hour | | Day | 24 GPU-hours | $110.64 | $4.61/hour | | Week | 168 GPU-hours | $774.48 | $4.61/hour | | Month | 720 GPU-hours | $3,319.20 | $4.61/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. - Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. - Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min · XMR ≈ 20 min ## B200 price questions ### What is the cheapest B200 cloud right now? On 29 October 2026 the lowest self-serve on-demand B200 price we found was $5.43, for Massed Compute's 8-GPU node, and GPUwithCrypto charges $4.61. Interruptible B200s cost less: Nebius preemptible from $0.99, CoreWeave spot at $4.26 and Verda spot at $4.39. They can be taken back mid-job. ### How much does a B200 cost per month? 720 GPU-hours cost $3,319.20 on GPUwithCrypto. At the median on-demand price of 29 October 2026 the same month cost $6,120.00, and $3,911.40 at Massed Compute's rate. ### Can a 70B model run on one B200? Yes, even in BF16. Llama 3.1 70B needs about 168 GB in BF16, within the B200's 180 GB. In FP8 it needs about 84 GB, which leaves room for two full 128K-token contexts. ### Can I rent a B200 with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID, no selfie and no card. Of the 9 self-serve B200 providers we checked, 3 accept crypto: Massed Compute (BTC), Spheron (USDC) and RunPod, which asks for KYC before the first crypto payment. The [crypto GPU cloud comparison](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") lists coins and rules. ### Can I rent a single B200, or only 8-GPU nodes? Both exist. Lambda rents 1, 2, 4 or 8 B200s, RunPod and Verda list single GPUs, and Massed Compute lists only an 8-GPU node. On GPUwithCrypto the GPU count depends on the host, and the dashboard shows it before you start. ## Sources 1. [Massed Compute — VM pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Hyperbolic — GPU marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29) 3. [Lambda — Pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 4. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 5. [Nebius — Prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 6. [CoreWeave — Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 7. [Verda — Pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 8. [Spheron — Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 9. [AWS — EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — Amazon Web Services (read 2026-10-29) 10. [Google Cloud — Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 11. [Crusoe — Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29) 12. [Vultr — Bare-metal plans API](https://api.vultr.com/v2/plans-metal?per_page=500) — Vultr (read 2026-10-29) 13. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 14. [Shadeform — Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29) 15. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 16. [NVIDIA HGX platform](https://www.nvidia.com/en-us/data-center/hgx/) — NVIDIA (read 2026-10-29) 17. [NVIDIA DGX B200](https://www.nvidia.com/en-us/data-center/dgx-b200/) — NVIDIA (read 2026-10-29) 18. [Introduction to the NVIDIA DGX B200 system](https://docs.nvidia.com/dgx/dgxb200-user-guide/introduction-to-dgxb200.html) — NVIDIA (read 2026-10-29) 19. [NVIDIA H200 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/h200/) — NVIDIA (read 2026-10-29) 20. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does it cost to rent an H100 per hour? > How much does it cost to rent an H100 per hour? $2.73 to $12.29 on demand at 16 providers on 29 Oct 2026. Ours: at least 15% under $2.73, in crypto, no KYC. - URL: https://gpuwithcrypto.com/gpu-prices/h100/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Every on-demand H100 price we found, with its spot or marketplace price, billing step, crypto option and ID rule. No ID and no card: top up in crypto, start an H100 as soon as the top-up is credited. **Quick answer — How much does it cost to rent an H100 per hour?** As of 29 October 2026, renting an on-demand NVIDIA H100 80GB costs $2.73 to $12.29 per GPU-hour across 16 providers. The median is $4.16. On **GPUwithCrypto** an H100 costs $2.32/hour, billed per minute and paid in crypto with no KYC. Spot and marketplace H100s go lower, from $0.79, but they can be interrupted. ## How much does an H100 cost per hour at each provider? One row per provider at its cheapest on-demand H100 configuration, sorted by price. According to GPUwithCrypto's price check of 29 October 2026, the most expensive on-demand H100 costs 4.5 times the cheapest. Azure charges $12.29 per GPU in an 8-GPU VM; Massed Compute charges $2.73. | Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC | | --- | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** | H100 80GB, 1 to 8 GPUs; SXM or PCIe shown per host | **$2.32** | — (on-demand only) | Per minute | Yes | No | | Massed Compute | H100 80GB (form factor not stated), 1 GPU | $2.73 | — | Hourly rate (increment not stated) | Yes (BTC) | None documented | | RunPod | H100 PCIe, 1 GPU (H100 SXM: $3.99) | $2.89 | $1.99 (Community Cloud) | Per second | Yes | KYC before first crypto payment | | OVHcloud | H100 80GB (form factor not stated), 1 GPU | $2.99 | — | Per second, 1-minute minimum | No | Sometimes | | Hyperbolic | H100 SXM | $3.19 | — | — | Unknown | None documented | | Thunder Compute | H100 SXM, 1 GPU | $3.20 | — | Per minute | No | None documented | | Lambda | H100 PCIe, 1 GPU (H100 SXM: $3.99 to $4.29) | $3.29 | — | Per minute | No (cards only) | None documented | | Jarvislabs | H100 SXM, 1 GPU | $3.49 | $1.54 (spot) | Per minute | No | None documented | | Crusoe | H100 SXM (HGX) | $3.90 | Contact sales | Per hour or per second | No | None documented | | DigitalOcean (Paperspace) | H100 SXM (HGX), 1 or 8 GPUs | $4.41 | — | Per second, 5-minute minimum | Yes (stablecoins, limited) | None documented | | Nebius | H100 SXM (HGX) | $4.50 | From $0.79 (preemptible) | — | No | None documented | | Verda (formerly DataCrunch) | H100 SXM, 1 GPU | $4.75 | $2.37 (spot) | — | No | None documented | | Spheron | H100 80GB (form factor not stated) | $5.86 | $2.79 (spot) | Per minute, 20-minute minimum | Yes (USDC) | None documented | | CoreWeave | H100 SXM, 8-GPU instance ÷ 8 | $6.16 | $2.46 (spot) | — | Unknown | Unknown | | AWS | H100, 1 GPU (p5.4xlarge) or 8 GPUs (p5.48xlarge) | $6.88 | — | — | No | None documented | | Google Cloud | H100 SXM, 8-GPU VM ÷ 8 (a3-highgpu-8g) | $11.06 | $6.59 (spot) | Per second, 1-minute minimum | No | None documented | | Microsoft Azure | H100, 8-GPU VM ÷ 8 (ND H100 v5) | $12.29 | $2.27 (spot) | — | No | None documented | | Vast.ai | H100 SXM, marketplace | — | $1.73 (Vast.ai's public price feed, cheapest rentable offer) | Per second | Yes (BitPay, Crypto.com) | None documented | | Akash Network | H100 SXM, decentralized marketplace | — | $2.04 (lowest recent provider bid) | — | Yes (AKT, ACT) | None documented | | Shadeform | H100 PCIe, 1 GPU, resold from other clouds | — | $2.50 (cheapest available listing) | — | Yes (USDC, per Stripe) | None documented | | Voltage Park | H100 SXM (HGX) | Excluded: FAQ says 'starting at' $1.99, pricing cards say 'Contact for pricing' | — | — | Unknown | Unknown | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded, rounded to the cent. Hyperscaler prices cover the whole VM (AWS US East, Azure East US, Google Cloud Iowa). 'None documented' means no ID step in the official sign-up, billing or terms pages we read, not a promise that a provider never asks. RunPod's Community Cloud is community-hosted. Spheron's own H100 page said 'from $2.65/hr' the same day; its pricing-page figure is shown. ## Why are spot and marketplace H100s cheaper than ours? Because they can be interrupted, or because an independent host sets the price and runs the machine. Our price is set against a named on-demand reference, not against these. - **Preemptible:** Nebius lists H100s from $0.79 per GPU-hour. Nebius can reclaim them, and the rate floats, capped one cent below its on-demand price. - **Spot:** Jarvislabs sells interruptible H100 SXM hours at $1.54. - **Marketplace:** Vast.ai's public price feed showed 56 rentable H100 SXM offers, from $1.73 with a median of $2.40. The feed does not say which offers are interruptible. - **Community-hosted:** RunPod's Community Cloud listed an H100 PCIe at $1.99, against $2.89 on its Secure Cloud. A job that checkpoints every few minutes can save money on spot. A long training run or a live inference server needs a GPU that stays put. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. ## Which H100 sets the reference price, and why is Voltage Park's $1.99 left out? The reference is Massed Compute's 'H100 (80GB)' at $2.73 per GPU-hour, and its page does not state the form factor. It could be SXM or PCIe. If you need an SXM card, compare with SXM prices: across 12 self-serve providers the lowest was $3.14 (Massed Compute) and the median $3.99. Voltage Park is left out because its on-demand pricing cards say 'Contact for pricing'. The $1.99 figure is a 'starting at' line in its FAQ, not a price you can pay without talking to sales. GPUwithCrypto has one H100 80GB price for every host. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## How much does an H100 cost per day, week or month? Our price for one GPU next to the lowest and the median self-serve on-demand price of 29 October 2026, for the same hours. | Period | GPU-hours | GPUwithCrypto | Lowest elsewhere (Massed Compute) | Median elsewhere | | --- | --- | --- | --- | --- | | 1 hour | 1 | **$2.32** | $2.73 | $4.16 | | 1 day | 24 | **$55.68** | $65.52 | $99.72 | | 1 week | 168 | **$389.76** | $458.64 | $698.04 | | 30 days | 720 | **$1,670.40** | $1,965.60 | $2,991.60 | *Prices checked on October 29, 2026.* The median is $4.155 per GPU-hour before rounding. Our day, week and month match the 24, 168 and 720 GPU-hour blocks. ## What are the H100's specs? NVIDIA's official figures for the three H100 versions sold in the cloud. Compare dense with dense: NVIDIA prints H100 Tensor figures with sparsity only, which doubles peak throughput for 2:4 sparse weights. | Spec | H100 SXM | H100 PCIe | H100 NVL | | --- | --- | --- | --- | | Memory | 80 GB HBM3 | 80 GB HBM2e | 94 GB HBM3 | | Memory bandwidth | 3,350 GB/s | 2,000 GB/s | 3,900 GB/s | | FP8 Tensor, dense | 1,979 TFLOPS (3,958 sparse) | Not on NVIDIA's current page | 1,670.5 TFLOPS (3,341 sparse) | | FP16 / BF16 Tensor, dense | 989.5 TFLOPS (1,979 sparse) | Not on NVIDIA's current page | 835.5 TFLOPS (1,671 sparse) | | GPU-to-GPU link | NVLink 900 GB/s | NVLink bridge 600 GB/s | NVLink bridge 600 GB/s | | Max power | 700 W | 350 W | 400 W | Dense = NVIDIA's sparse figure ÷ 2. Almost no off-the-shelf LLM checkpoint is 2:4 sparse, so the dense figure matches real training and inference. ## What fits on one H100? Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb) and weights plus optimizer and gradients for fine-tuning. A full 128K-token context for Llama 3.1 70B adds another 42.9 GB of BF16 KV cache. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the formulas. | Model and job | Memory needed | One H100 80GB? | | --- | --- | --- | | Llama 3.1 8B inference, BF16 | ≈ 19.2 GB | Yes | | Llama 3.1 8B LoRA fine-tune | ≈ 19.2 GB | Yes | | Llama 3.1 8B full fine-tune | ≈ 128 to 144 GB before activations | No: 2× 80 GB or more | | Llama 3.1 70B inference, 4-bit | ≈ 42 GB | Yes | | Llama 3.1 70B QLoRA fine-tune | ≈ 52.5 GB | Yes | | Llama 3.1 70B inference, FP8 | ≈ 84 GB | No: just over 80 GB; one H100 NVL (94 GB) or one H200 | | Llama 3.1 70B inference, BF16 | ≈ 168 GB | No: 3× H100 | | Qwen2.5 72B inference, 4-bit | ≈ 43.6 GB | Yes | | FLUX.1 [dev] transformer, BF16 weights | 24 GB, plus text encoders and VAE | Yes | | DeepSeek-V3 or R1 inference, FP8 | ≈ 805 GB | No: 11× 80 GB | ## Which H100 providers accept crypto? Of the 16 self-serve H100 providers above, 4 accept crypto: Massed Compute (BTC), RunPod, DigitalOcean (stablecoins, limited availability) and Spheron (USDC). RunPod's billing docs say to complete any required KYC before the first crypto payment. The marketplaces Vast.ai and Akash also take crypto. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR, from a $1 top-up, with no ID, no selfie and no card. The [H100 with crypto page](https://gpuwithcrypto.com/rent-h100-with-crypto/ "Where can I rent an H100 and pay with crypto?") compares the crypto options step by step. ## What does an H100 cost on GPUwithCrypto? $2.32/hour, at least 15% under its named on-demand reference: Massed Compute at $2.73, read on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $23.20 | $2.32/hour | | Day | 24 GPU-hours | $55.68 | $2.32/hour | | Week | 168 GPU-hours | $389.76 | $2.32/hour | | Month | 720 GPU-hours | $1,670.40 | $2.32/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. - Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. - Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min · XMR ≈ 20 min ## H100 price questions ### What is the cheapest H100 cloud right now? On 29 October 2026 the lowest self-serve on-demand H100 price we found was $2.73 at Massed Compute, and GPUwithCrypto charges $2.32. Interruptible H100s cost less: Nebius preemptible from $0.79, and $1.73 in Vast.ai's public price feed. ### Is an H100 SXM more expensive than an H100 PCIe? Usually. The median H100 SXM cost $3.99 per GPU-hour across 12 self-serve providers, while the two PCIe prices were $2.89 (RunPod) and $3.29 (Lambda). The SXM has 3,350 GB/s of memory bandwidth against 2,000 GB/s. ### How much does an H100 cost per month? 720 GPU-hours cost $1,670.40 on GPUwithCrypto. At the median on-demand price of 29 October 2026 the same month cost $2,991.60, and $1,965.60 at the lowest. ### Can I rent an H100 with crypto and no KYC? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID, no selfie and no card. Elsewhere, 4 of the 16 self-serve H100 providers we checked accept crypto, and RunPod asks for KYC before the first crypto payment. ### Should I rent an H100 or an A100? Rent the A100 when your model fits in 80 GB and runs in BF16: it costs $0.92/hour here. Rent the H100 for FP8, which the A100 lacks, and for up to 64% more memory bandwidth on the SXM version. The [H100 vs A100 comparison](https://gpuwithcrypto.com/compare/h100-vs-a100/ "H100 vs A100: which should I rent?") goes further. ## Sources 1. [Massed Compute — VM pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 3. [OVHcloud — Public Cloud prices](https://www.ovhcloud.com/en/public-cloud/prices/) — OVHcloud (read 2026-10-29) 4. [Hyperbolic — GPU marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29) 5. [Thunder Compute — Pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 6. [Lambda — Pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 7. [Jarvislabs — Pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 8. [Crusoe — Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29) 9. [DigitalOcean — GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 10. [Nebius — Prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 11. [Verda — Pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 12. [Spheron — Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 13. [CoreWeave — Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 14. [AWS — EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — Amazon Web Services (read 2026-10-29) 15. [Google Cloud — Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 16. [Azure Retail Prices API, East US virtual machines](https://prices.azure.com/api/retail/prices?$filter=armRegionName%20eq%20'eastus'%20and%20serviceName%20eq%20'Virtual%20Machines') — Microsoft (read 2026-10-29) 17. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 18. [Akash Network — GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) 19. [Shadeform — Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29) 20. [Voltage Park — Pricing](https://www.voltagepark.com/pricing) — Voltage Park (read 2026-10-29) 21. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 22. [NVIDIA H100 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/h100/) — NVIDIA (read 2026-10-29) 23. [NVIDIA H100 PCIe product brief](https://www.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf) — NVIDIA (read 2026-10-29) 24. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does an H200 cost per hour? > How much does an H200 cost per hour? $3.62 to $10.60 on demand across 13 providers on 29 Oct 2026. Ours: at least 15% under $3.62, in crypto, no KYC. - URL: https://gpuwithcrypto.com/gpu-prices/h200/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Every on-demand H200 price we found, spot and marketplace prices, what fits in 141 GB, and who takes crypto or asks for ID. No ID and no card: top up in crypto, start an H200 as soon as the top-up is credited. **Quick answer — How much does an H200 cost per hour?** As of 29 October 2026, an NVIDIA H200 141GB costs from $3.62 to $10.60 per GPU-hour on demand across 13 providers, with a median of $5.29. On **GPUwithCrypto** it costs $3.07/hour, billed per minute and paid in crypto with no KYC. Interruptible H200s start lower: Nebius lists preemptible H200s from $0.79. ## How much does an H200 cost per hour at each provider? One row per provider at its cheapest on-demand H200 configuration, sorted by price. The most expensive on-demand H200 we found (Google Cloud, $10.60 per GPU in an 8-GPU VM) costs 2.9 times the cheapest. | Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC | | --- | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** | H200 141GB, 1 to 8 GPUs; form factor shown per host | **$3.07** | — (on-demand only) | Per minute | Yes | No | | Massed Compute | H200 141GB NVL, 1 GPU | $3.62 | — | Hourly rate (increment not stated) | Yes (BTC) | None documented | | Hyperbolic | H200 141GB (form factor not stated) | $3.99 | — | — | Unknown | None documented | | Crusoe | H200 SXM (HGX) | $4.29 | Contact sales | Per hour or per second | No | None documented | | DigitalOcean (Paperspace) | H200 SXM (HGX), 1 or 8 GPUs | $4.47 | — | Per second, 5-minute minimum | Yes (stablecoins, limited) | None documented | | Jarvislabs | H200 SXM, 1 GPU | $4.59 | $2.29 (spot) | Per minute | No | None documented | | Spheron | H200 141GB (form factor not stated) | $4.80 | $3.35 (spot) | Per minute, 20-minute minimum | Yes (USDC) | None documented | | RunPod | H200 141GB (form factor not stated), 1 GPU | $5.29 | $3.59 (Community Cloud) | Per second | Yes | KYC before first crypto payment | | Nebius | H200 SXM (HGX) | $5.40 | From $0.79 (preemptible) | — | No | None documented | | Verda (formerly DataCrunch) | H200 SXM, 1 GPU | $6.19 | $3.09 (spot) | — | No | None documented | | OVHcloud | H200 141GB (form factor not stated), 8-GPU instance ÷ 8 | $6.20 | — | Per second, 1-minute minimum | No | Sometimes | | CoreWeave | H200 SXM, 8-GPU instance ÷ 8 | $6.31 | $2.62 (spot) | — | Unknown | Unknown | | AWS | H200, 8 GPUs (p5en.48xlarge) ÷ 8 | $7.91 | — | — | No | None documented | | Google Cloud | H200 SXM, 8-GPU VM ÷ 8 (a3-ultragpu-8g) | $10.60 | $6.36 (spot) | Per second, 1-minute minimum | No | None documented | | Vast.ai | H200 NVL and H200, marketplace | — | $3.00 for an H200 NVL, $3.77 for an H200 (Vast.ai's public price feed, cheapest rentable offers) | Per second | Yes (BitPay, Crypto.com) | None documented | | Shadeform | H200 SXM, 8 GPUs, resold from other clouds | — | $4.00 (cheapest available listing) | — | Yes (USDC, per Stripe) | None documented | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded, rounded to the cent. Hyperscaler prices cover the whole VM (AWS US East, Google Cloud Iowa). 'None documented' means no ID step in the official sign-up, billing or terms pages we read, not a promise that a provider never asks. RunPod's Community Cloud is community-hosted. ## Is the H200 worth the extra cost over an H100? Yes when your model does not fit in 80 GB. According to GPUwithCrypto's price check of 29 October 2026, a median on-demand H200 costs 27% more per hour than a median H100 ($5.29 against $4.16). It has 76% more memory (141 GB against 80 GB) and 43% more bandwidth than the H100 SXM (4,800 against 3,350 GB/s). Compute is the same: NVIDIA gives the H200 SXM and the H100 SXM the same FP8 rating, 3,958 TFLOPS with sparsity or about 1,979 dense. So the H200 pays off for big models and long contexts, not for small ones. Llama 3.1 70B in FP8 needs about 84 GB: one H200, or two H100s. The [H100 vs H200 comparison](https://gpuwithcrypto.com/compare/h100-vs-h200/ "H100 vs H200: is the H200 worth the extra cost?") has the full verdict, and the [H100 price page](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") the H100 rows. ## Why are some marketplace and spot H200s cheaper than ours? Because the host sets the price on a marketplace, and the provider can take a spot GPU back. On 29 October 2026, the cheapest rentable H200 NVL offer in Vast.ai's public price feed was $3.00 per hour, below our $3.07. Nebius listed preemptible H200s from $0.79 and Jarvislabs an H200 spot at $2.29. These prices suit jobs that save checkpoints and can restart. For a long run or an always-on inference server, compare on-demand prices. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. ## Which H200 sets the reference price: SXM or NVL? The reference is Massed Compute's H200 NVL at $3.62 per GPU-hour. The NVL is the PCIe version: same 141 GB and 4,800 GB/s as the SXM, but 600 W and a lower Tensor rating. The cheapest self-serve H200 SXM we found was Crusoe's at $4.29. GPUwithCrypto has one H200 price for every host. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## How much does an H200 cost per day, week or month? Our price for one GPU next to the lowest and the median self-serve on-demand price of 29 October 2026, for the same number of hours. | Period | GPU-hours | GPUwithCrypto | Lowest elsewhere (Massed Compute) | Median elsewhere | | --- | --- | --- | --- | --- | | 1 hour | 1 | **$3.07** | $3.62 | $5.29 | | 1 day | 24 | **$73.68** | $86.88 | $126.96 | | 1 week | 168 | **$515.76** | $608.16 | $888.72 | | 30 days | 720 | **$2,210.40** | $2,606.40 | $3,808.80 | *Prices checked on October 29, 2026.* Our day, week and month match the 24, 168 and 720 GPU-hour blocks. ## What are the H200's specs? NVIDIA's official figures for the two H200 versions. NVIDIA prints Tensor figures with sparsity only; the dense figures below are half of those, so compare dense with dense. | Spec | H200 SXM | H200 NVL (PCIe) | | --- | --- | --- | | Memory | 141 GB HBM3e | 141 GB HBM3e | | Memory bandwidth | 4,800 GB/s | 4,800 GB/s | | FP8 Tensor, dense | 1,979 TFLOPS (3,958 sparse) | 1,670.5 TFLOPS (3,341 sparse) | | FP16 / BF16 Tensor, dense | 989.5 TFLOPS (1,979 sparse) | 835.5 TFLOPS (1,671 sparse) | | GPU-to-GPU link | NVLink 900 GB/s | NVLink bridge 900 GB/s | | Max power | 700 W | 600 W | NVIDIA labels the H200 figures 'Preliminary specifications. May be subject to change.' ## What fits on one H200? Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb). The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the formulas and more models. | Model and job | Memory needed | One H200 141GB? | | --- | --- | --- | | Llama 3.1 70B inference, FP8 | ≈ 84 GB | Yes | | Llama 3.1 70B, FP8 weights plus one full 128K-token context (BF16 cache) | ≈ 113 GB | Yes | | Qwen2.5 72B inference, FP8 | ≈ 87.2 GB | Yes | | Llama 3.1 70B QLoRA fine-tune | ≈ 52.5 GB | Yes | | Llama 3.1 70B inference, BF16 | ≈ 168 GB | No: 2× H200 | | Llama 3.1 70B LoRA fine-tune | ≈ 140 to 168 GB | No: 2× H200 | | Llama 3.1 405B inference, FP8 | ≈ 486 GB | No: 4× H200 | | DeepSeek-V3 or R1 inference, FP8 | ≈ 805 GB | No: one 8× H200 node (1,128 GB) | ## What does an H200 cost on GPUwithCrypto? $3.07/hour, at least 15% under its named on-demand reference: Massed Compute at $3.62, read on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $30.70 | $3.07/hour | | Day | 24 GPU-hours | $73.68 | $3.07/hour | | Week | 168 GPU-hours | $515.76 | $3.07/hour | | Month | 720 GPU-hours | $2,210.40 | $3.07/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. - Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. - Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min · XMR ≈ 20 min ## H200 price questions ### What is the cheapest H200 cloud right now? On 29 October 2026 the lowest self-serve on-demand H200 price we found was $3.62, for Massed Compute's H200 NVL, and GPUwithCrypto charges $3.07. The cheapest H200 SXM was Crusoe's at $4.29. Spot H200s go lower but can be interrupted, and marketplace hosts set their own prices. ### How much does an H200 cost per month? 720 GPU-hours cost $2,210.40 on GPUwithCrypto. At the median on-demand price of 29 October 2026 the same month cost $3,808.80, and $2,606.40 at the lowest. ### Can a 70B model run on one H200? Yes in FP8 or 4-bit. Llama 3.1 70B needs about 84 GB in FP8, which leaves room for long contexts on a 141 GB H200. In BF16 it needs about 168 GB, so two H200s. ### Can I rent an H200 with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID, no selfie and no card. Of the 13 self-serve H200 providers we checked, 4 accept crypto: Massed Compute, DigitalOcean, Spheron and RunPod. RunPod asks for KYC before the first crypto payment. ### Which H200 do I get on GPUwithCrypto, SXM or NVL? It depends on the host. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. Both versions have 141 GB of HBM3e and 4,800 GB/s. ## Sources 1. [Massed Compute — VM pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Hyperbolic — GPU marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29) 3. [Crusoe — Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29) 4. [DigitalOcean — GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 5. [Jarvislabs — Pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 6. [Spheron — Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 7. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 8. [Nebius — Prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 9. [Verda — Pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 10. [OVHcloud — Public Cloud prices](https://www.ovhcloud.com/en/public-cloud/prices/) — OVHcloud (read 2026-10-29) 11. [CoreWeave — Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 12. [AWS — EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — Amazon Web Services (read 2026-10-29) 13. [Google Cloud — Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 14. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 15. [Shadeform — Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29) 16. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 17. [NVIDIA H200 Tensor Core GPU](https://www.nvidia.com/en-us/data-center/h200/) — NVIDIA (read 2026-10-29) 18. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does an L40S cost per hour? > How much does an L40S cost per hour? $0.97 to $3.15 on demand across 11 providers on 29 Oct 2026. Ours: at least 15% under $0.97, in crypto, no KYC. - URL: https://gpuwithcrypto.com/gpu-prices/l40s/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Every on-demand L40S price we found, spot and marketplace prices, how it compares with the A100, and what fits in 48 GB. No ID and no card: top up in crypto, start an L40S as soon as the top-up is credited. **Quick answer — How much does an L40S cost per hour?** As of 29 October 2026, an NVIDIA L40S 48GB costs from $0.97 to $3.15 per GPU-hour on demand across 11 providers, with a median of $1.67. On **GPUwithCrypto** it costs $0.82/hour, billed per minute and paid in crypto with no KYC. Marketplace L40S offers go lower: $0.33 in Vast.ai's public price feed. ## How much does an L40S cost per hour at each provider? One row per provider at its cheapest on-demand L40S configuration, sorted by price. Seven of the 11 providers charge between $1.50 and $1.89; the most expensive costs 3.2 times the cheapest. | Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC | | --- | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** | L40S 48GB, 1 to 8 GPUs | **$0.82** | — (on-demand only) | Per minute | Yes | No | | Massed Compute | L40S, 1 GPU | $0.97 | — | Hourly rate (increment not stated) | Yes (BTC) | None documented | | RunPod | L40S, 1 GPU | $1.09 | $0.79 (Community Cloud) | Per second | Yes | KYC before first crypto payment | | Crusoe | L40S | $1.50 | Contact sales | Per hour or per second | No | None documented | | Nebius | L40S, Intel host ('from' price; AMD host $1.82) | $1.55 | From $0.74 (preemptible) | — | No | None documented | | DigitalOcean (Paperspace) | L40S, 1 GPU | $1.57 | — | Per second, 5-minute minimum | Yes (stablecoins, limited) | None documented | | Vultr | L40S, 1 GPU (2 and 4 GPUs at the same rate) | $1.67 | — | Hourly | Yes (BitPay) | None documented | | OVHcloud | L40S, 1 GPU | $1.80 | — | Per second, 1-minute minimum | No | Sometimes | | AWS | L40S, 1 GPU (g6e.xlarge) | $1.86 | — | — | No | None documented | | Verda (formerly DataCrunch) | L40S, 1 GPU | $1.88 | $0.94 (spot) | — | No | None documented | | CoreWeave | L40S, 8-GPU instance ÷ 8 | $2.25 | $0.99 (spot) | — | Unknown | Unknown | | Spheron | L40S | $3.15 | $1.18 (spot) | Per minute, 20-minute minimum | Yes (USDC) | None documented | | Vast.ai | L40S, marketplace | — | $0.33 (Vast.ai's public price feed, cheapest rentable offer) | Per second | Yes (BitPay, Crypto.com) | None documented | | Shadeform | L40S, 2 GPUs, resold from other clouds | — | $1.50 (cheapest available listing) | — | Yes (USDC, per Stripe) | None documented | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded, rounded to the cent. The AWS price covers the whole instance (US East). Vultr's figure comes from its public plans API, which did not confirm availability. 'None documented' means no ID step in the official pages we read, not a promise that a provider never asks. ## L40S vs A100: which is the better deal per hour? For FP8 inference and image generation, the L40S. According to GPUwithCrypto's price check of 29 October 2026, a median on-demand L40S costs 25% less than a median A100 80GB ($1.67 against $2.23). It also has FP8 Tensor Cores, which the A100 lacks. The A100 keeps two advantages: 80 GB of memory against 48 GB, and 2,039 GB/s of bandwidth on the SXM version against 864 GB/s. A job that needs more than 48 GB, such as a 70B model in 4-bit with a 32K-token context (about 53 GB), needs the A100. On GPUwithCrypto the two cost $0.82/hour and $0.92/hour. The [L40S vs A100 comparison](https://gpuwithcrypto.com/compare/l40s-vs-a100/ "L40S vs A100 for inference and fine-tuning: which should I rent?") goes further. ## Why are marketplace and spot L40S cheaper than ours? Because a host sets the marketplace price, and the provider can stop a spot GPU. On 29 October 2026, Vast.ai's public price feed listed 61 rentable L40S offers, from $0.33 with a median of $0.73. Nebius listed preemptible L40S from $0.74, and RunPod's community-hosted Community Cloud charged $0.79. Those three prices were below our $0.82 that day. They fit batch jobs that can restart. For a server that must stay up, compare on-demand prices: ours is 51% below the on-demand median. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. ## How much does an L40S cost per day, week or month? Our price for one GPU next to the lowest and the median self-serve on-demand price of 29 October 2026, for the same number of hours. | Period | GPU-hours | GPUwithCrypto | Lowest elsewhere (Massed Compute) | Median elsewhere | | --- | --- | --- | --- | --- | | 1 hour | 1 | **$0.82** | $0.97 | $1.67 | | 1 day | 24 | **$19.68** | $23.28 | $40.10 | | 1 week | 168 | **$137.76** | $162.96 | $280.73 | | 30 days | 720 | **$590.40** | $698.40 | $1,203.12 | *Prices checked on October 29, 2026.* The median is $1.671 per GPU-hour before rounding. Our day, week and month match the 24, 168 and 720 GPU-hour blocks. ## What are the L40S's specs? | Attribute | Value | | --- | --- | | Architecture | Ada Lovelace, PCIe Gen4, dual slot | | Memory | 48 GB GDDR6 | | Memory bandwidth | 864 GB/s | | FP8 Tensor | 733 TFLOPS dense, 1,466 with sparsity | | FP16 / BF16 Tensor | 362.05 TFLOPS dense, 733 with sparsity | | FP32 | 91.6 TFLOPS | | NVLink | No | | Max power | 350 W | ## What fits on one L40S? Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb) and weights plus adapters for LoRA. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the formulas. | Model and job | Memory needed | One L40S 48GB? | | --- | --- | --- | | Llama 3.1 8B inference, BF16 | ≈ 19.2 GB | Yes | | Llama 3.1 8B LoRA fine-tune | ≈ 19.2 GB | Yes | | Llama 3.1 70B inference, 4-bit | ≈ 42 GB | Yes | | Qwen2.5 72B inference, 4-bit | ≈ 43.6 GB | Yes | | Llama 3.1 70B QLoRA fine-tune | ≈ 52.5 GB | Borderline: the QLoRA authors fine-tuned a 65B model on one 48 GB GPU | | FLUX.1 [dev] transformer, BF16 weights | 24 GB, plus text encoders and VAE | Yes | | FLUX.1 [dev] LoRA training | ≈ 28.8 GB | Yes | | SDXL 1.0 base, FP16 weights | 7 GB | Yes | | Llama 3.1 70B inference, FP8 | ≈ 84 GB | No: 2× L40S | | Llama 3.1 8B full fine-tune | ≈ 128 to 144 GB before activations | No | ## What does an L40S cost on GPUwithCrypto? $0.82/hour, at least 15% under its named on-demand reference: Massed Compute at $0.97, read on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $8.20 | $0.82/hour | | Day | 24 GPU-hours | $19.68 | $0.82/hour | | Week | 168 GPU-hours | $137.76 | $0.82/hour | | Month | 720 GPU-hours | $590.40 | $0.82/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. - Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. - Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min · XMR ≈ 20 min ## L40S price questions ### What is the cheapest L40S cloud? On 29 October 2026 the lowest self-serve on-demand L40S price we found was $0.97 at Massed Compute, and GPUwithCrypto charges $0.82. Vast.ai's public price feed showed marketplace offers from $0.33, set by independent hosts. ### How much does an L40S cost per month? 720 GPU-hours cost $590.40 on GPUwithCrypto. At the median on-demand price of 29 October 2026 the same month cost $1,203.12, and $698.40 at the lowest. ### Can an L40S run a 70B model? Yes in 4-bit: Llama 3.1 70B needs about 42 GB and fits in 48 GB. In FP8 it needs about 84 GB, so two L40S. The L40S has no NVLink, so multi-GPU runs go over PCIe. ### Is the L40S good for ComfyUI and Flux? It has the memory for it. FLUX.1 [dev] weights take 24 GB in BF16, and LoRA training needs about 28.8 GB, both inside 48 GB. See the [best GPU for ComfyUI and Flux](https://gpuwithcrypto.com/best-gpu-for/comfyui-flux/ "What is the best cloud GPU for ComfyUI, Stable Diffusion and Flux?"). ### Can I rent an L40S with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID and no card. Of the 11 self-serve L40S providers we checked, 5 accept crypto: Massed Compute, DigitalOcean, Vultr, Spheron and RunPod. RunPod asks for KYC before the first crypto payment. ## Sources 1. [Massed Compute — VM pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 3. [Crusoe — Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29) 4. [Nebius — Prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 5. [DigitalOcean — GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 6. [Vultr — Public plans API](https://api.vultr.com/v2/plans?type=all&per_page=500) — Vultr (read 2026-10-29) 7. [OVHcloud — Public Cloud prices](https://www.ovhcloud.com/en/public-cloud/prices/) — OVHcloud (read 2026-10-29) 8. [AWS — EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — Amazon Web Services (read 2026-10-29) 9. [Verda — Pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 10. [CoreWeave — Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 11. [Spheron — Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 12. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 13. [Shadeform — Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29) 14. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 15. [NVIDIA L40S GPU](https://www.nvidia.com/en-us/data-center/l40s/) — NVIDIA (read 2026-10-29) 16. [QLoRA: Efficient Finetuning of Quantized LLMs](https://arxiv.org/abs/2305.14314) — Dettmers et al. (read 2026-10-29) 17. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does it cost to rent an RTX 4090 per hour? > How much does it cost to rent an RTX 4090 per hour? $0.33 to $0.89 on demand on 29 Oct 2026, Salad rising to $0.44 on 1 Nov. Less here, in crypto, no KYC. - URL: https://gpuwithcrypto.com/gpu-prices/rtx-4090/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) The on-demand RTX 4090 prices we found, the cheaper marketplace and interruptible offers, Salad's 1 November price rise, and what fits in 24 GB. No ID and no card: top up in crypto, start an RTX 4090 as soon as the top-up is credited. **Quick answer — How much does it cost to rent an RTX 4090 per hour?** As of 29 October 2026, renting an on-demand NVIDIA RTX 4090 costs $0.33 per hour at Salad and $0.89 at RunPod. They are the only two providers with a current self-serve on-demand price. Salad's price rises to $0.44 on 1 November. On **GPUwithCrypto** an RTX 4090 costs $0.28/hour, billed per minute and paid in crypto with no KYC. ## How much does an RTX 4090 cost per hour at each provider? According to GPUwithCrypto's price check of 29 October 2026, only 2 of the 24 providers we priced list a current self-serve on-demand RTX 4090. RunPod's Secure Cloud price is 2.7 times Salad's High Priority price. Marketplace and interruptible offers fill the rest. | Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC | | --- | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** | RTX 4090 24GB, whole GPUs | **$0.28** | — (on-demand only) | Per minute | Yes | No | | Salad | RTX 4090, High Priority tier, distributed consumer GPUs | $0.33 until 31 Oct; $0.44 from 1 Nov | $0.16 (Batch priority, most interruptible) | Per second | Yes (USDC, USDT, RENDER on Solana) | None documented | | RunPod | RTX 4090, 1 GPU, Secure Cloud | $0.89 | $0.34 (Community Cloud) | Per second | Yes | KYC before first crypto payment | | Spheron | RTX 4090 | Excluded: $0.72 marked 'last known Sep 17' | — | Per minute, 20-minute minimum | Yes (USDC) | None documented | | Vast.ai | RTX 4090, marketplace | — | $0.13 (Vast.ai's public price feed, cheapest rentable offer) | Per second | Yes (BitPay, Crypto.com) | None documented | | Akash Network | RTX 4090, decentralized marketplace | — | $0.34 (recent provider bids) | — | Yes (AKT, ACT) | None documented | | TensorDock | RTX 4090, marketplace | — | $0.35 'typical' (page dated July 24, 2024; not counted) | Per second | Unknown | None documented | | Shadeform | RTX 4090, 8 GPUs, resold from other clouds | — | None available | — | Yes (USDC, per Stripe) | None documented | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded. Salad's figures come from its official price-change post of 26 October 2026; salad.com/pricing already shows the 1 November prices. RunPod's Community Cloud is community-hosted. 'None documented' means no ID step in the official pages we read. ## What changes on 1 November 2026? Salad's High Priority RTX 4090 goes from $0.33 to $0.44 per hour, a 33% rise announced in its price-change post of 26 October 2026. High Priority is Salad's tier that is never preempted, and it set our reference price. Our price was computed from the $0.33 figure, so it stays below Salad's new rate. Salad's cheaper tiers remain: from 1 November, Medium costs $0.347, Low $0.253 and Batch $0.160. Those tiers can be preempted. ## Why are marketplace RTX 4090s cheaper than ours? Because most of them run on independent hosts' machines, at prices the hosts set, and many can be interrupted. On 29 October 2026, Vast.ai's public price feed listed 562 rentable RTX 4090 offers, from $0.13 with a median of $0.49. Salad's Batch tier, its most interruptible, cost $0.16. Both were below our $0.28. They suit image batches and LoRA runs that can restart. For a ComfyUI server or a training run that must not stop, compare on-demand prices. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. The [spot GPU guide](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?") explains the trade-off. ## How much does an RTX 4090 cost per day, week or month? Our price for one GPU next to Salad's High Priority price before and after 1 November 2026, and the median of 29 October. | Period | GPU-hours | GPUwithCrypto | Salad until 31 Oct ($0.33) | Salad from 1 Nov ($0.44) | Median elsewhere ($0.61) | | --- | --- | --- | --- | --- | --- | | 1 hour | 1 | **$0.28** | $0.33 | $0.44 | $0.61 | | 1 day | 24 | **$6.72** | $7.92 | $10.56 | $14.64 | | 1 week | 168 | **$47.04** | $55.44 | $73.92 | $102.48 | | 30 days | 720 | **$201.60** | $237.60 | $316.80 | $439.20 | *Prices checked on October 29, 2026.* Our day, week and month match the 24, 168 and 720 GPU-hour blocks, and unused GPU-hours never expire. Salad credits expire 12 months after purchase, per its billing docs. ## What are the RTX 4090's specs? | Attribute | Value | | --- | --- | | Architecture | Ada Lovelace, PCIe Gen4 desktop card | | Memory | 24 GB GDDR6X | | Memory bandwidth | 1,008 GB/s | | FP8 Tensor (FP32 accumulate) | 330.3 TFLOPS dense, 660.6 with sparsity | | FP16 / BF16 Tensor (FP32 accumulate) | 165.2 TFLOPS dense, 330.4 with sparsity | | FP32 | 82.6 TFLOPS | | NVLink | No | | Total graphics power | 450 W | | Marketing 'AI TOPS' | 1,321: sparse FP8 with FP16 accumulate, not comparable with dense data-center figures | ## What fits on one RTX 4090? Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb). Image pipelines also load text encoders and a VAE. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the formulas. | Model and job | Memory needed | One RTX 4090 24GB? | | --- | --- | --- | | Llama 3.1 8B inference, BF16 | ≈ 19.2 GB | Yes | | Mistral 7B inference, BF16 | ≈ 17.5 GB | Yes | | Llama 3.1 8B LoRA fine-tune | ≈ 19.2 GB | Yes | | Llama 3.1 8B QLoRA fine-tune | ≈ 6 GB | Yes | | SDXL 1.0 base, FP16 weights | 7 GB | Yes | | FLUX.1 [dev] transformer, FP8 weights | 12 GB, plus text encoders and VAE | Yes | | FLUX.1 [dev] transformer, BF16 weights | 24 GB | Tight: the weights alone fill the card | | FLUX.1 [dev] LoRA training, BF16 base | ≈ 28.8 GB | No; QLoRA (≈ 9 GB) fits | | Llama 3.1 70B inference, 4-bit | ≈ 42 GB | No | ## What does an RTX 4090 cost on GPUwithCrypto? $0.28/hour, at least 15% under its named on-demand reference: Salad at $0.33, read on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $2.80 | $0.28/hour | | Day | 24 GPU-hours | $6.72 | $0.28/hour | | Week | 168 GPU-hours | $47.04 | $0.28/hour | | Month | 720 GPU-hours | $201.60 | $0.28/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. - Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. - Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min · XMR ≈ 20 min ## RTX 4090 rental questions ### What is the cheapest RTX 4090 cloud? Among on-demand RTX 4090s that are never preempted, the lowest list price on 29 October 2026 was Salad's High Priority tier at $0.33, which becomes $0.44 on 1 November. GPUwithCrypto's $0.28/hour is at least 15% under that reference. Marketplace offers go lower: Vast.ai's public price feed started at $0.13. ### How much does an RTX 4090 cost per month? 720 GPU-hours cost $201.60 on GPUwithCrypto. At Salad's High Priority price from 1 November 2026 the same month costs $316.80, and $439.20 at the median of 29 October. ### Is an RTX 4090 enough for Flux and ComfyUI? For inference, yes with FP8 weights: FLUX.1 [dev] takes 12 GB in FP8 and 24 GB in BF16. LoRA training in BF16 needs about 28.8 GB, so use QLoRA or a 32 GB RTX 5090. See the [best GPU for ComfyUI and Flux](https://gpuwithcrypto.com/best-gpu-for/comfyui-flux/ "What is the best cloud GPU for ComfyUI, Stable Diffusion and Flux?"). ### RTX 4090 or RTX 5090: which should I rent? The RTX 5090 has 32 GB against 24 GB and 1,792 GB/s of bandwidth against 1,008 GB/s. On GPUwithCrypto it costs $0.42/hour against $0.28/hour. Pick it when 24 GB is not enough. Prices are on the [RTX 5090 page](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?"). ### Can I rent an RTX 4090 with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID and no card. Salad takes USDC, USDT and RENDER on Solana, Vast.ai uses BitPay and Crypto.com, and RunPod asks for KYC before the first crypto payment. ## Sources 1. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 2. [SaladCloud — Billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 3. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 4. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 5. [Spheron — Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 6. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 7. [Akash Network — GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) 8. [TensorDock — Cloud GPUs](https://www.tensordock.com/cloud-gpus.html) — TensorDock (read 2026-10-29) 9. [Shadeform — Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29) 10. [NVIDIA GeForce RTX 4090](https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4090/) — NVIDIA (read 2026-10-29) 11. [NVIDIA Ada GPU architecture whitepaper](https://images.nvidia.com/aem-dam/Solutions/geforce/ada/nvidia-ada-gpu-architecture.pdf) — NVIDIA (read 2026-10-29) 12. [FLUX.1 [dev] model card](https://huggingface.co/black-forest-labs/FLUX.1-dev) — Black Forest Labs (read 2026-10-29) 13. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does it cost to rent an RTX 5090 in the cloud? > How much does it cost to rent an RTX 5090? $0.50 to $1.19 an hour on demand on 29 Oct 2026, Salad rising to $0.60 on 1 Nov. Less here, in crypto, no KYC. - URL: https://gpuwithcrypto.com/gpu-prices/rtx-5090/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) The three on-demand RTX 5090 prices we found, the cheaper marketplace and interruptible offers, Salad's 1 November price rise, and what fits in 32 GB. No ID and no card: top up in crypto, start an RTX 5090 as soon as the top-up is credited. **Quick answer — How much does it cost to rent an RTX 5090 in the cloud?** As of 29 October 2026, renting an NVIDIA RTX 5090 on demand costs $0.50 per hour at Salad, $0.84 at Spheron and $1.19 at RunPod. Salad's price rises to $0.60 on 1 November. On **GPUwithCrypto** an RTX 5090 costs $0.42/hour, billed per minute and paid in crypto with no KYC. Marketplace offers in Vast.ai's public price feed start lower, at $0.39. ## How much does an RTX 5090 cost per hour at each provider? Every RTX 5090 price we found on 29 October 2026, on-demand first. All three self-serve on-demand providers accept crypto; only RunPod asks for KYC first. | Provider | Configuration | On-demand $/GPU-hour | Spot / marketplace $/GPU-hour | Billing increment | Accepts crypto | ID / KYC | | --- | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** | RTX 5090 32GB, whole GPUs | **$0.42** | — (on-demand only) | Per minute | Yes | No | | Salad | RTX 5090, High Priority tier, distributed consumer GPUs | $0.50 until 31 Oct; $0.60 from 1 Nov | $0.25 (Batch priority, most interruptible) | Per second | Yes (USDC, USDT, RENDER on Solana) | None documented | | Spheron | RTX 5090, price marked 'last known Oct 28' | $0.84 | — | Per minute, 20-minute minimum | Yes (USDC) | None documented | | RunPod | RTX 5090, 1 GPU, Secure Cloud | $1.19 | $0.69 (Community Cloud) | Per second | Yes | KYC before first crypto payment | | Vast.ai | RTX 5090, marketplace | — | $0.39 (Vast.ai's public price feed, cheapest rentable offer) | Per second | Yes (BitPay, Crypto.com) | None documented | | Akash Network | RTX 5090, decentralized marketplace | — | $0.57 (recent provider bids; none available) | — | Yes (AKT, ACT) | None documented | | Shadeform | RTX 5090, 8 GPUs, resold from other clouds | — | None available | — | Yes (USDC, per Stripe) | None documented | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded. Salad's figures come from its official price-change post of 26 October 2026; salad.com/pricing already shows the 1 November prices. RunPod's Community Cloud is community-hosted. 'None documented' means no ID step in the official pages we read. ## Is an RTX 5090 worth the extra cost over an RTX 4090? Yes when 24 GB is not enough. According to GPUwithCrypto's price check of 29 October 2026, the median on-demand RTX 5090 costs 38% more per hour than the median RTX 4090. That is $0.84 against $0.61. For that it brings 33% more memory (32 GB against 24 GB) and 78% more memory bandwidth (1,792 against 1,008 GB/s). The extra 8 GB is what matters for image work. FLUX.1 [dev] LoRA training needs about 28.8 GB with a BF16 base: it fits a 5090, not a 4090. On GPUwithCrypto the two cost $0.42/hour and $0.28/hour. See the [RTX 4090 prices](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?"). ## What changes on 1 November 2026? Salad's High Priority RTX 5090 goes from $0.50 to $0.60 per hour, a 20% rise announced in its price-change post of 26 October 2026. High Priority is Salad's tier that is never preempted, and it set our reference price. Our price was computed from the $0.50 figure, so it stays below Salad's new rate. Salad's cheaper tiers remain: from 1 November, Medium costs $0.483, Low $0.367 and Batch $0.250. Those tiers can be preempted. ## Why are marketplace and interruptible RTX 5090s cheaper than ours? Because independent hosts set marketplace prices on their own machines, and interruptible tiers can be stopped. On 29 October 2026, Vast.ai's public price feed listed 970 rentable RTX 5090 offers, from $0.39 with a median of $0.73. Salad's Batch tier cost $0.25, and its Low tier from 1 November costs $0.367. All three are below our $0.42. For a job that can restart, they save money. For a server or a long training run, compare on-demand prices: ours is 50% below the on-demand median. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. ## How much does an RTX 5090 cost per day, week or month? Our price for one GPU next to Salad's High Priority price before and after 1 November 2026, and the median on-demand price of 29 October. | Period | GPU-hours | GPUwithCrypto | Salad until 31 Oct ($0.50) | Salad from 1 Nov ($0.60) | Median elsewhere ($0.84) | | --- | --- | --- | --- | --- | --- | | 1 hour | 1 | **$0.42** | $0.50 | $0.60 | $0.84 | | 1 day | 24 | **$10.08** | $12.00 | $14.40 | $20.16 | | 1 week | 168 | **$70.56** | $84.00 | $100.80 | $141.12 | | 30 days | 720 | **$302.40** | $360.00 | $432.00 | $604.80 | *Prices checked on October 29, 2026.* Our day, week and month match the 24, 168 and 720 GPU-hour blocks, and unused GPU-hours never expire. Salad credits expire 12 months after purchase, per its billing docs. ## What are the RTX 5090's specs? | Attribute | Value | | --- | --- | | Architecture | Blackwell, PCIe Gen5 desktop card | | Memory | 32 GB GDDR7 | | Memory bandwidth | 1,792 GB/s | | FP8 Tensor (FP32 accumulate) | 419 TFLOPS dense, 838 with sparsity | | FP16 / BF16 Tensor (FP32 accumulate) | 209.5 TFLOPS dense, 419 with sparsity | | FP4 Tensor | 1,676 TFLOPS dense, 3,352 with sparsity | | FP32 | 104.8 TFLOPS | | Total graphics power | 575 W | | Marketing 'AI TOPS' | 3,352: sparse FP4, not comparable with dense data-center figures | ## What fits on one RTX 5090? Memory needed for common open models, using weights × 1.2 for inference (EleutherAI's rule of thumb), before driver overhead. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the formulas. | Model and job | Memory needed | One RTX 5090 32GB? | | --- | --- | --- | | Llama 3.1 8B inference, BF16 | ≈ 19.2 GB | Yes | | Llama 3.1 8B, BF16 weights plus a full 128K-token context with FP8 cache | ≈ 24.6 GB | Yes | | Llama 3.1 8B LoRA fine-tune | ≈ 19.2 GB | Yes | | FLUX.1 [dev] transformer, BF16 weights | 24 GB, plus text encoders and VAE | Yes | | FLUX.1 [dev] LoRA training, BF16 base | ≈ 28.8 GB | Yes | | SDXL 1.0 base, FP16 weights | 7 GB | Yes | | Llama 3.1 70B inference, 4-bit | ≈ 42 GB | No | | Llama 3.1 8B full fine-tune | ≈ 128 to 144 GB before activations | No | ## What does an RTX 5090 cost on GPUwithCrypto? $0.42/hour, at least 15% under its named on-demand reference: Salad at $0.50, read on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $4.20 | $0.42/hour | | Day | 24 GPU-hours | $10.08 | $0.42/hour | | Week | 168 GPU-hours | $70.56 | $0.42/hour | | Month | 720 GPU-hours | $302.40 | $0.42/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. - Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. - Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min · XMR ≈ 20 min ## RTX 5090 rental questions ### What is the cheapest RTX 5090 cloud? Among on-demand RTX 5090s that are never preempted, the lowest list price on 29 October 2026 was Salad's High Priority tier at $0.50, which becomes $0.60 on 1 November. GPUwithCrypto's $0.42/hour is at least 15% under that reference. Marketplace offers in Vast.ai's public price feed started at $0.39. ### How much does an RTX 5090 cost per month? 720 GPU-hours cost $302.40 on GPUwithCrypto. At Salad's High Priority price from 1 November 2026 the same month costs $432.00, and $604.80 at the median of 29 October. ### RTX 5090 vs H100: which should I rent for AI? The H100 has 80 GB and NVLink; the RTX 5090 has 32 GB. Rent the 5090 for image models and LLMs up to about 8B in BF16, the H100 for bigger models. On GPUwithCrypto they cost $0.42/hour and $2.32/hour. See [RTX 5090 vs H100](https://gpuwithcrypto.com/compare/rtx-5090-vs-h100/ "RTX 5090 vs H100 for AI: when is the cheap card enough?"). ### Can I train a Flux LoRA on an RTX 5090? Yes. FLUX.1 [dev] LoRA training with a BF16 base needs about 28.8 GB, which fits in 32 GB. On a 24 GB RTX 4090 you would need QLoRA instead. ### Can I rent an RTX 5090 with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID and no card. Salad, Spheron and RunPod also accept crypto, but RunPod asks for KYC before the first crypto payment. ## Sources 1. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 2. [SaladCloud — Billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 3. [Spheron — Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 4. [Spheron — Billing](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 5. [Runpod — Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 6. [Runpod — Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 7. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 8. [Akash Network — GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) 9. [Shadeform — Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29) 10. [NVIDIA GeForce RTX 5090](https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/) — NVIDIA (read 2026-10-29) 11. [NVIDIA RTX Blackwell GPU architecture whitepaper](https://images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdf) — NVIDIA (read 2026-10-29) 12. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How do I fine-tune Llama on a rented GPU? > How to fine-tune Llama on a rented GPU with QLoRA or LoRA: TRL commands, 6 GB for Llama 3.1 8B in QLoRA, and saving the adapter before deleting. Paid in crypto. - URL: https://gpuwithcrypto.com/guides/fine-tune-llama-on-rented-gpu/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) A LoRA or QLoRA adapter is the cheapest way to teach Llama your data. This guide picks the GPU, runs the training with Hugging Face TRL, and gets the adapter off the machine before the disk is wiped. **Quick answer — How do I fine-tune Llama on a rented GPU?** To fine-tune Llama on a rented GPU, train a LoRA or QLoRA adapter with Hugging Face TRL. Llama 3.1 8B needs about 6 GB in QLoRA and 19.2 GB in LoRA, so one 24 GB RTX 4090 is enough; Llama 3.1 70B in QLoRA needs about 52.5 GB, one A100 80GB. Copy the adapter off before deleting the instance. As of 29 October 2026, **GPUwithCrypto** rents the RTX 4090 at $0.28/hour, paid in crypto. ## Which GPU do I need to fine-tune Llama? Memory by the rules in our [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?"): QLoRA ≈ 0.75 GB per billion parameters, LoRA = 2 bytes per parameter × 1.2, full fine-tuning 16–18 bytes per parameter. Our prices of 29 October 2026. | Model | Method | VRAM (rule) | GPU | GPUwithCrypto per hour | | --- | --- | --- | --- | --- | | Llama 3.1 8B | QLoRA | 6 GB | 1× RTX 4090 (24 GB) | $0.28/hour | | Llama 3.1 8B | LoRA (BF16 base) | 19.2 GB | 1× RTX 4090 (24 GB) | $0.28/hour | | Llama 3.1 8B | Full fine-tune | 128–144 GB + activations | 1× B200 (180 GB) | $4.61/hour | | Llama 3.1 70B | QLoRA | 52.5 GB | 1× A100 80GB or 1× H100 | $0.92/hour or $2.32/hour | | Llama 3.1 70B | LoRA (BF16 base) | 168 GB | 1× B200 or 2× H200 | $4.61/hour or $6.14/hour | *Prices checked on October 29, 2026.* Disk: the Llama 3.1 8B Instruct weights take 16.1 GB, the 70B Instruct weights 141.1 GB. Leave room for checkpoints. ## How do I run a QLoRA fine-tune of Llama 3.1 8B? Commands from the Hugging Face TRL and Hub documentation. Swap the model id for `meta-llama/Llama-3.1-70B-Instruct` on an 80 GB GPU to train the 70B model the same way. New to rented GPUs? [How to rent a GPU for AI](https://gpuwithcrypto.com/guides/rent-a-gpu-for-ai/ "How do I rent a GPU for AI, step by step?") covers the account, SSH and a first job. 1. **Rent the GPU** — Start a 1× RTX 4090 instance for the 8B model, or a 1× A100 80GB for 70B. Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. 2. **Get access to Llama** — The meta-llama repositories on Hugging Face are gated: request access on the model page and accept Meta's licence, then run hf auth login on the instance with a read token. 3. **Install TRL, PEFT and bitsandbytes** — pip install "trl[peft]" bitsandbytes. bitsandbytes provides the 4-bit loading that QLoRA needs. 4. **Pick a dataset** — Use a Hugging Face dataset id or your own file of conversations. The example uses trl-lib/Capybara, the dataset in TRL's own examples. 5. **Train the adapter** — Run the trl sft command below inside tmux. --load_in_4bit loads the base in 4-bit, --use_peft adds the LoRA adapter, and checkpoints land in --output_dir. 6. **Test the adapter** — Load the base model, attach the adapter with PEFT and compare answers with the base model on your own prompts. 7. **Copy the adapter off, then delete** — Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. Copy the output folder to your computer or push it to a private Hugging Face repository, then delete the instance. *train-qlora.sh* ```bash pip install "trl[peft]" bitsandbytes hf auth login trl sft \ --model_name_or_path meta-llama/Llama-3.1-8B-Instruct \ --dataset_name trl-lib/Capybara \ --load_in_4bit \ --use_peft \ --lora_r 32 \ --lora_alpha 16 \ --lora_target_modules q_proj k_proj v_proj o_proj gate_proj up_proj down_proj \ --learning_rate 2.0e-4 \ --num_train_epochs 1 \ --per_device_train_batch_size 1 \ --gradient_accumulation_steps 16 \ --gradient_checkpointing \ --save_strategy steps \ --save_steps 200 \ --output_dir llama-3.1-8b-qlora ``` ## What do the main training flags do? - `--lora_r 32 --lora_alpha 16`: the adapter's rank and scaling, the values in TRL's PEFT examples. Rank 32 on all seven linear layers of Llama 3.1 8B trains about 84 million parameters, about 1% of the model. - `--learning_rate 2.0e-4`: TRL's documentation recommends about 10 times the full fine-tuning rate (2.0e-5) for LoRA. - `--per_device_train_batch_size 1 --gradient_accumulation_steps 16`: an effective batch of 16 with the memory of a batch of 1. - `--gradient_checkpointing`: recomputes activations instead of storing them, trading time for memory. - `--save_strategy steps --save_steps 200`: writes a checkpoint every 200 steps, so a crash costs at most 200 steps. *save-adapter.sh* ```bash # from your computer: copy the adapter folder before you delete the instance rsync -avP -e "ssh -p SSH_PORT" root@HOST:/root/llama-3.1-8b-qlora/ ./llama-3.1-8b-qlora/ # or, on the instance: create a private repo on huggingface.co first, then upload hf upload your-name/llama-3.1-8b-qlora ./llama-3.1-8b-qlora . # later, serve the base model with the adapter in vLLM vllm serve meta-llama/Llama-3.1-8B-Instruct \ --enable-lora --max-lora-rank 32 \ --lora-modules my-adapter=./llama-3.1-8b-qlora ``` ## What does fine-tuning Llama on a rented GPU cost? Cost = GPUs × wall-clock hours × price per GPU-hour. **The hours below are assumptions for illustration, not measurements**: run time depends on your dataset's tokens, sequence length and batch size. We chose hours that match our blocks of 10 and 24 GPU-hours. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. Every GPU and block is listed on [GPU pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). A small GPU is the biggest saving. According to GPUwithCrypto's price check of 29 October 2026, the lowest self-serve on-demand RTX 4090 cost $0.33 per hour (Salad) and the lowest H100 $2.73 (Massed Compute): 8.3 times more, for an 8B QLoRA job that fits in 24 GB either way. The [RTX 4090 price page](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") lists the providers we checked. ### Run-cost estimate under assumed hours | Job | Configuration | Assumed hours (not measured) | GPU-hours | GPUwithCrypto cost | | --- | --- | --- | --- | --- | | Llama 3.1 8B, QLoRA | 1× RTX 4090 | 10 | 10 | $2.80 | | Llama 3.1 8B, LoRA | 1× RTX 4090 | 24 | 24 | $6.72 | | Llama 3.1 70B, QLoRA | 1× A100 80GB | 24 | 24 | $22.08 | | Llama 3.1 70B, LoRA | 2× H200 | 12 | 24 | $73.68 | *Prices checked on October 29, 2026.* ## How do I avoid losing work on a rented GPU? Save often, copy off early and watch the balance. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. When the GPU-hours of a GPU run out, its instances stop and their disks are kept; a new block of hours lets you start them again. If a run stops, Hugging Face's Trainer resumes from the last checkpoint in the output folder when you call `trainer.train(resume_from_checkpoint=True)` from Python. Unsloth and Axolotl are two other well-known LoRA tools; this guide uses TRL because its command-line flags are documented on Hugging Face's own pages. ## Llama fine-tuning questions ### How much VRAM do I need to fine-tune Llama 3.1 8B? About 6 GB with QLoRA and 19.2 GB with LoRA on a BF16 base, so a 24 GB RTX 4090 runs both. Full fine-tuning needs 128 to 144 GB before activations, one 180 GB B200. ### Can I fine-tune Llama 70B on one GPU? Yes, with QLoRA on an 80 GB A100 or H100: the estimate is 52.5 GB. LoRA on a BF16 base needs about 168 GB, one B200 or two H200s. More in [the best GPU to fine-tune Llama 70B](https://gpuwithcrypto.com/best-gpu-for/fine-tuning-llama-70b/ "What is the best GPU to fine-tune Llama 70B (LoRA, QLoRA, full)?"). ### Should I use LoRA or QLoRA? QLoRA when memory is the limit: it loads the base model in 4-bit and roughly divides the memory by three against LoRA. LoRA when the GPU has room: the base stays in BF16. Compare both adapters on your own evaluation set. ### How long does fine-tuning Llama take? It depends on the number of training tokens, the sequence length and the batch size, and we have not measured a run, so we give no timing. A 10-hour block is enough to measure your steps per hour before a long run. ### Can I pay for the GPU with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR from a $1 top-up, with no ID and no card. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. ## Sources 1. [TRL command line interfaces](https://huggingface.co/docs/trl/clis) — Hugging Face (read 2026-10-29) 2. [TRL PEFT integration (LoRA, QLoRA, saving adapters)](https://huggingface.co/docs/trl/peft_integration) — Hugging Face (read 2026-10-29) 3. [Hugging Face CLI guide (hf auth login, hf upload)](https://huggingface.co/docs/huggingface_hub/guides/cli) — Hugging Face (read 2026-10-29) 4. [Trainer and TrainingArguments](https://huggingface.co/docs/transformers/main_classes/trainer) — Hugging Face (read 2026-10-29) 5. [Llama 3.1 8B model card](https://huggingface.co/meta-llama/Llama-3.1-8B) — Meta / Hugging Face (read 2026-10-29) 6. [Llama 3.1 70B model card](https://huggingface.co/meta-llama/Llama-3.1-70B) — Meta / Hugging Face (read 2026-10-29) 7. [The Llama 3 Herd of Models (Table 3)](https://arxiv.org/abs/2407.21783) — arXiv (read 2026-10-29) 8. [QLoRA: Efficient Finetuning of Quantized LLMs](https://arxiv.org/abs/2305.14314) — arXiv (read 2026-10-29) 9. [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) — arXiv (read 2026-10-29) 10. [vLLM recipes (LoRA serving flags)](https://recipes.vllm.ai/) — vLLM project (read 2026-10-29) 11. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 12. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How do I rent a GPU for AI, step by step? > How to rent a GPU for AI step by step: pick it by VRAM, top up in crypto, start an instance, SSH in, save outputs. 24 GB to 180 GB GPUs, no KYC, October 2026. - URL: https://gpuwithcrypto.com/guides/rent-a-gpu-for-ai/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Renting a cloud GPU takes a few minutes once you know which one you need. These are the steps with GPUwithCrypto, then a checklist of what to compare at any provider. **Quick answer — How do I rent a GPU for AI, step by step?** To rent a GPU for AI, pick the GPU whose VRAM holds your model plus about 20%, open an account, add credit, start an instance, connect over SSH, and copy your results off before you delete it. With **GPUwithCrypto** the account is a number, credit is topped up in crypto from $1, and an H100 80GB costs $2.32/hour as of 29 October 2026, billed per minute. ## Renting a GPU with GPUwithCrypto at a glance | Attribute | Value | | --- | --- | | GPUs | NVIDIA H100 80GB, H200 141GB, B200, A100 80GB, L40S 48GB, GeForce RTX 4090 24GB and RTX 5090 32GB, rented by the minute | | Price rule | Each GPU is priced at least 15% under a named reference: the on-demand price of the same GPU at a self-serve provider, read on 29 October 2026 (H100: Massed Compute at $2.73 per GPU-hour; each GPU's reference is named on its price page). Spot, interruptible, marketplace and contact-sales prices are not used as references, and some of them are lower. We do not claim to be the cheapest for every GPU | | Billing | Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once | | Payment | Crypto only: BTC, USDT, USDC, ETH, SOL, LTC, XMR, minimum top-up $1 | | KYC | None: no ID, no selfie, no card | | Instance | Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image | | How to try a GPU | There is no free tier: a 10-hour block is the smallest purchase to try a GPU, and unused hours never expire | ## How do I rent a GPU for AI with GPUwithCrypto? 1. **Choose the GPU by VRAM** — Add 20% to the model's weights (parameters × bytes per parameter) and pick the smallest GPU that holds it: a 24 GB RTX 4090 for 7–8B models, a 48 GB L40S for 32B in FP8, a 141 GB H200 for 70B in FP8. The VRAM guide has the full table. 2. **Get an account number** — Sign in with an account number; an email address is optional. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good. 3. **Top up in crypto** — Send at least $1 in BTC, USDT, USDC, ETH, SOL, LTC, XMR, on the exact network shown on the payment page. The balance is credited after the network confirmations. 4. **Add your SSH key** — You add your SSH public key in the dashboard; we never see or store your private key. 5. **Start an instance** — Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. Stock changes through the day: the dashboard shows which GPUs are free right now, and nothing is charged until an instance starts. Once your payment is confirmed, you launch an instance from the dashboard; our team prepares it at the GPU provider and its SSH details appear there as soon as it runs. Billing starts only then. 6. **Connect over SSH and run your job** — Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. Run nvidia-smi first to check the GPU model and its memory, and start long jobs inside tmux so a dropped connection does not stop them. 7. **Save your outputs, then stop or delete** — Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. *first-session.sh* ```bash # from your computer: connect (host and SSH port are shown in the dashboard) ssh -p SSH_PORT root@HOST # on the instance: GPU model, memory and driver version nvidia-smi # run long jobs inside tmux (install it with: apt-get install -y tmux) tmux new -s job # from your computer: copy results back before you delete the instance rsync -avP -e "ssh -p SSH_PORT" root@HOST:/root/outputs/ ./outputs/ ``` ## What should I compare when renting a GPU anywhere? Compare like with like, then read the rules around the price. The eight points that change the real cost: 1. **Price type.** On-demand, spot, community and marketplace prices are different products. Spot and marketplace GPUs can be taken back; compare on-demand with on-demand. See [what is a spot GPU](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?"). 2. **Billing increment and minimum.** On 29 October 2026, RunPod billed per second, Lambda and Thunder Compute per minute, DigitalOcean per second with a 5-minute minimum, and Spheron per minute after a 20-minute minimum runtime. 3. **Form factor.** An H100 SXM moves 3,350 GB/s of memory bandwidth; an H100 PCIe 2,000 GB/s. Check which one the price is for. 4. **Whole GPU or slice.** A MIG partition or a time-shared GPU is not the same product as a whole card. 5. **Storage.** Disks are often billed apart: RunPod charged $0.10 per GB per month for container disk. Check whether the disk survives a stop and what happens on delete. 6. **Payment and ID.** Whether crypto is accepted, whether paying in crypto triggers KYC, and whether a card is required. 7. **Refunds.** Several providers never refund credit bought with crypto. 8. **Acceptable use.** Read what the provider forbids before you build on it. ## Which GPU clouds take crypto, and do they ask for ID? From each provider's own billing docs and terms, read on 29 October 2026. "No ID step documented" means the pages we read describe none, not a promise that the provider never asks. Full list of 33 providers: [GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?"). | Provider | Accepts crypto | ID / KYC | Card needed | Credit bought with crypto | | --- | --- | --- | --- | --- | | **GPUwithCrypto** | Yes: BTC, USDT, USDC, ETH, SOL, LTC, XMR | No | No | Never expires; hours of a failed instance go back to the balance | | RunPod | Yes (Crypto.com named in its terms) | KYC before the first crypto payment | Auto-pay is card-only | Credits non-refundable | | Vast.ai | Yes (BitPay, Crypto.com) | No ID step documented | Verified credit card per its billing FAQ | Not refunded | | Salad | Yes (USDC, USDT, RENDER on Solana) | No ID step documented | — | Not refundable; credits expire after 12 months | | Massed Compute | Yes (BTC) | No ID step documented | — | No auto-recharge for crypto | | Spheron | Yes (USDC on six chains) | No ID step documented | — | No self-serve refunds; credits do not expire | | Lambda | No: "only accepts major credit cards" | No ID step documented | Credit card (debit and prepaid declined) | — | *Prices checked on October 29, 2026.* ## How many GPU clouds accept crypto? Fewer than half. According to GPUwithCrypto's check of 29 October 2026, 14 of the 33 GPU providers we checked accept crypto, and 4 of the 33 ask for ID in some cases. Only RunPod ties a KYC step to paying in crypto, in its billing docs. None documents an ID check for every customer at sign-up. More: [no-KYC GPU cloud](https://gpuwithcrypto.com/no-kyc-gpu-cloud/ "Is there a GPU cloud with no KYC?") and [renting a GPU without a credit card](https://gpuwithcrypto.com/rent-gpu-without-credit-card/ "How can I rent a GPU without a credit card?"). ## How do I keep the bill down? Right-size the GPU, stop it when idle and test before a long run. A model that fits a 48 GB L40S does not need an 80 GB H100. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once, so stop the instance between sessions. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. Before a multi-day job, run it for an hour on the same GPU and measure your steps per hour. All prices: [pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## Questions about renting a GPU ### Can I rent a GPU without a credit card? Yes. GPUwithCrypto takes only crypto (BTC, USDT, USDC, ETH, SOL, LTC, XMR) and asks for no card and no ID. At other providers, crypto sometimes comes with conditions: RunPod asks for KYC before the first crypto payment, and Vast.ai's billing FAQ asks for a verified card. ### How fast does a rented GPU start? Once your payment is confirmed, you launch an instance from the dashboard; our team prepares it at the GPU provider and its SSH details appear there as soon as it runs. Billing starts only then. The top-up is credited once the network has confirmed the transaction; the usual time per coin is shown on the payment page. ### What happens when my GPU-hours run out? When the GPU-hours of a GPU run out, its instances stop and their disks are kept; a new block of hours lets you start them again. Copy your outputs off before the hours run out, or buy another block. ### Is my data deleted when I delete the instance? Yes. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. For billing we record each instance’s GPU model, host, start and stop time; these lines are kept for 30 days. We do not read the files on your disk. Account number, orders and top-ups stay while the account exists, then only as long as accounting law asks. Authorities receive records only when a law compels us. ### What can I not do with a rented GPU? No cryptocurrency mining, no password or hash cracking, no attacks on other systems (DDoS, scanning, exploitation), no malware, no spam, no child sexual abuse material, no intimate or deceptive images of real people without their consent, no weapons development and no work for sanctioned people or companies. The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") has the details. ## Sources 1. [RunPod billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 2. [RunPod pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 3. [Vast.ai billing reference](https://docs.vast.ai/documentation/reference/billing) — Vast.ai (read 2026-10-29) 4. [SaladCloud billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 5. [Massed Compute crypto payments](https://vm-docs.massedcompute.com/docs/billing/crypto/automatic-payments) — Massed Compute (read 2026-10-29) 6. [Spheron billing](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 7. [Lambda billing](https://docs.lambda.ai/public-cloud/manage-billing/) — Lambda (read 2026-10-29) 8. [Lambda pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 9. [DigitalOcean GPU Droplet pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 10. [Spheron pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 11. [NVIDIA H100 specifications](https://www.nvidia.com/en-us/data-center/h100/) — NVIDIA (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How do I run DeepSeek V4 on a cloud GPU? > How to run DeepSeek V4 on a cloud GPU: V4 Flash (284B, 159.6 GB of weights) on 4× H200, V4 Pro (1.6T, 864.7 GB) on 8× H200, with vLLM or SGLang. Crypto, no KYC. - URL: https://gpuwithcrypto.com/guides/run-deepseek-v4-on-cloud-gpu/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) DeepSeek V4 comes in two open-weight sizes, Flash and Pro. This guide gives the VRAM arithmetic for each, the GPU counts the serving engines document, the launch commands and the hourly cost. **Quick answer — How do I run DeepSeek V4 on a cloud GPU?** To run DeepSeek V4 on a cloud GPU, rent enough VRAM for its weights plus about 20%: V4 Flash (284B parameters, 159.6 GB of weights) needs at least two H200s, and the SGLang cookbook documents four; V4 Pro (1.6T, 864.7 GB) needs eight H200s. Serve it with vLLM or SGLang. As of 29 October 2026, **GPUwithCrypto** rents the H200 at $3.07/hour per GPU, paid in crypto. ## DeepSeek V4 at a glance (Hugging Face, read 29 October 2026) | Attribute | Value | | --- | --- | | Models | DeepSeek-V4-Flash and DeepSeek-V4-Pro, open weights under the MIT License | | Parameters | Flash: 284B total, 13B activated per token · Pro: 1.6T total, 49B activated | | Weight format | Mixed: MoE expert weights in FP4, most other weights in FP8 | | Weights on disk | Flash 159.6 GB (46 files) · Pro 864.7 GB (64 files) | | Context length | 1 million tokens; DeepSeek recommends at least 384K for the Think Max mode | | Serving engines on the model card | vLLM and SGLang | | Sampling advice on the card | temperature 1.0, top_p 1.0 | ## How much VRAM does DeepSeek V4 need? Flash needs about 191.5 GB and Pro about 1,037.6 GB: the published weights × 1.2. DeepSeek V4 is a mixture-of-experts model. Only 13B (Flash) or 49B (Pro) parameters work on each token, which cuts compute, not memory: every expert must sit in VRAM. Its compressed attention keeps the KV cache small, so the usual per-token cache formula for Llama-style models does not apply. The weights × 1.2 rule for other models is on [how much VRAM do I need?](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") ### Which GPUs hold DeepSeek V4? Formula minimum = the fewest GPUs whose combined VRAM holds weights × 1.2. Documented layouts = what the vLLM recipes and the SGLang cookbook describe. Our price per hour for the whole configuration, 29 October 2026. | Model | Weights | × 1.2 | Formula minimum | Documented layouts | GPUwithCrypto per hour | | --- | --- | --- | --- | --- | --- | | V4 Flash | 159.6 GB | 191.5 GB | 2× H200 (282 GB) or 2× B200 (360 GB) | SGLang: 4 GPUs on H200 or B200, 8 on H100 · vLLM: 4 GPUs per replica on H200 or B200 | $6.14/hour; $12.28/hour for the documented layout | | V4 Pro | 864.7 GB | 1,037.6 GB | 8× H200 (1,128 GB) or 8× B200 (1,440 GB) | SGLang: 8 GPUs on H200 or B200; H100 needs two 8-GPU nodes · vLLM: 8 GPUs on H200, context capped at 800K tokens | $24.56/hour | *Prices checked on October 29, 2026.* 8× H100 (640 GB) is too small for V4 Pro, and our hosts have at most 8 GPUs. The [H100 vs H200](https://gpuwithcrypto.com/compare/h100-vs-h200/ "H100 vs H200: is the H200 worth the extra cost?") page compares their memory and price. H100 and H200 have no FP4 Tensor Cores: SGLang runs the FP4 experts on them with W4A16 kernels, while the B200 runs FP4 natively. ## How do I serve DeepSeek V4 Flash with vLLM? The layout below is the one the vLLM recipe recommends: four GPUs, data parallel plus expert parallel. Every flag comes from that recipe. 1. **Rent four H200s** — Start a 4× H200 instance (564 GB). Check that its disk holds the 159.6 GB of weights plus room for logs. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. 2. **Install vLLM** — Install the latest tagged vLLM release; the recipe asks for vLLM 0.20.0 or newer for this model. You can also start the instance from the vllm/vllm-openai Docker image. 3. **Download the weights** — hf download deepseek-ai/DeepSeek-V4-Flash saves the 46 weight files to the Hugging Face cache. vLLM also downloads them on first start. 4. **Start the server** — Run the command below. It binds to 127.0.0.1 so the API is reachable only through your SSH tunnel. 5. **Send a request** — Forward port 8000 over SSH and call the OpenAI-compatible endpoint /v1/chat/completions. 6. **Stop the instance when you are done** — Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. *serve-v4-flash-vllm.sh (4× H200)* ```bash pip install -U vllm hf download deepseek-ai/DeepSeek-V4-Flash vllm serve deepseek-ai/DeepSeek-V4-Flash \ --trust-remote-code \ --data-parallel-size 4 \ --enable-expert-parallel \ --kv-cache-dtype fp8 \ --block-size 256 \ --tokenizer-mode deepseek_v4 \ --reasoning-parser deepseek_v4 \ --tool-call-parser deepseek_v4 \ --enable-auto-tool-choice \ --host 127.0.0.1 --port 8000 ``` *request.sh (from your computer)* ```bash # forward the API port (host and SSH port are shown in the dashboard) ssh -N -L 8000:127.0.0.1:8000 -p SSH_PORT root@HOST # in another terminal curl http://127.0.0.1:8000/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{"model": "deepseek-ai/DeepSeek-V4-Flash", "messages": [{"role": "user", "content": "Explain the KV cache in two sentences."}]}' ``` ## How do I serve DeepSeek V4 with SGLang instead? Start the instance from the `lmsysorg/sglang:latest` Docker image, which the SGLang cookbook lists for H100, H200 and B200, and use tensor parallelism: `--tp 4` for Flash on four H200s, `--tp 8` for Flash on eight H100s or for Pro on eight H200s. The API answers on port 30000 with the same OpenAI-compatible routes. *serve-v4-flash-sglang.sh (4× H200)* ```bash python3 -m sglang.launch_server \ --model-path deepseek-ai/DeepSeek-V4-Flash \ --trust-remote-code \ --tp 4 \ --host 127.0.0.1 \ --port 30000 ``` ## How do I serve DeepSeek V4 Pro? On eight H200s, with data parallelism across all eight GPUs. The vLLM recipe for H200 caps the context at 800K tokens to leave room for the cache, because the dense weights are copied to every GPU. For the Think Max mode, keep at least 384K tokens of context, as DeepSeek recommends. *serve-v4-pro-vllm.sh (8× H200)* ```bash vllm serve deepseek-ai/DeepSeek-V4-Pro \ --trust-remote-code \ --data-parallel-size 8 \ --enable-expert-parallel \ --max-model-len 800000 \ --kv-cache-dtype fp8 \ --block-size 256 \ --tokenizer-mode deepseek_v4 \ --reasoning-parser deepseek_v4 \ --host 127.0.0.1 --port 8000 ``` ## How much does it cost to run DeepSeek V4 per hour? At our prices, V4 Flash on the documented four-H200 layout costs $12.28/hour, and V4 Pro on eight H200s costs $24.56/hour. A full day of one H200 is $73.68. Provider by provider: [H200 rental prices](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?"). Pro's weights are 5.4 times the size of Flash's (864.7 against 159.6 GB). According to GPUwithCrypto's price check of 29 October 2026, the lowest self-serve on-demand H200 price was $3.62 per GPU-hour (Massed Compute): $14.48 per hour for four GPUs and $28.96 for eight. Every price: [pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## Which DeepSeek V4 checkpoint should I download? Start with `deepseek-ai/DeepSeek-V4-Flash` or `deepseek-ai/DeepSeek-V4-Pro`, the repositories created on 22 April 2026, whose cards give the figures above. DeepSeek later published updated checkpoints, `DeepSeek-V4-Flash-0731` and `DeepSeek-V4-Pro-0813`, with a built-in draft head for speculative decoding; the vLLM recipe asks for vLLM 0.25.0 or newer for them. The `-Base` repositories are for further pre-training, not for chat or tool calls. ## DeepSeek V4 questions ### How much VRAM do I need for DeepSeek V4 Flash? About 191.5 GB: 159.6 GB of weights plus 20%. Two H200s (282 GB) hold it by that rule; the SGLang cookbook and the vLLM recipe both document four H200s, which leaves more room for the cache and for parallel users. ### How much VRAM do I need for DeepSeek V4 Pro? About 1,037.6 GB: 864.7 GB of weights plus 20%. Eight H200s (1,128 GB) or eight B200s (1,440 GB) hold it on one host. Eight H100s (640 GB) do not. ### Can I run DeepSeek V4 on H100s? V4 Flash, yes: the SGLang cookbook documents eight H100s with `--tp 8`. V4 Pro needs two 8-GPU H100 nodes in the cookbook, more than one of our instances holds, so use eight H200s. ### Can I run DeepSeek V4 on one GPU? Not on the GPUs we rent. The largest, a 180 GB B200, holds Flash's 159.6 GB of weights but leaves only 20 GB for the cache, below the 20% margin. Two GPUs is the floor. ### Can I pay for the H200s with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR from a $1 top-up, with no ID and no card. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. ## Sources 1. [DeepSeek-V4-Flash model card and files](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) — DeepSeek / Hugging Face (read 2026-10-29) 2. [DeepSeek-V4-Pro model card and files](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro) — DeepSeek / Hugging Face (read 2026-10-29) 3. [vLLM recipe: DeepSeek V4 Flash](https://recipes.vllm.ai/deepseek-ai/DeepSeek-V4-Flash) — vLLM project (read 2026-10-29) 4. [vLLM recipe: DeepSeek V4 Pro](https://recipes.vllm.ai/deepseek-ai/DeepSeek-V4-Pro) — vLLM project (read 2026-10-29) 5. [DeepSeek-V4 cookbook](https://docs.sglang.io/cookbook/autoregressive/DeepSeek/DeepSeek-V4) — SGLang (read 2026-10-29) 6. [Hugging Face CLI guide (hf download)](https://huggingface.co/docs/huggingface_hub/guides/cli) — Hugging Face (read 2026-10-29) 7. [NVIDIA H200 specifications](https://www.nvidia.com/en-us/data-center/h200/) — NVIDIA (read 2026-10-29) 8. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much VRAM do I need to run or fine-tune an LLM? > How much VRAM do I need for an LLM? Parameters × bytes + 20%: a 70B model needs 168 GB in FP16, 84 GB in FP8. Fit tables and crypto prices, October 2026. - URL: https://gpuwithcrypto.com/how-much-vram-do-i-need/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) The formula, the arithmetic for open models from 7B to 1.6T parameters, and the smallest set of our GPUs that holds each one, with its price per hour. **Quick answer — How much VRAM do I need to run or fine-tune an LLM?** The VRAM you need is the model's parameters × bytes per parameter, plus about 20% for the KV cache and activations: 2.4 GB per billion parameters in FP16/BF16, 1.2 GB in FP8, 0.6 GB in 4-bit. A 70B model needs about 168 GB, 84 GB or 42 GB. As of 29 October 2026, **GPUwithCrypto** rents the GPUs that fit, from an RTX 4090 at $0.28/hour to an H200 at $3.07/hour, paid in crypto. ## What is the formula? | Attribute | Value | | --- | --- | | Weights | Parameters (billions) × bytes per parameter = GB | | Bytes per parameter | FP16 / BF16 = 2 · FP8 / INT8 = 1 · 4-bit = 0.5 | | Running a model | Weights × 1.2. EleutherAI puts the extra memory for inference at no more than 20% of the weights | | Long context | Add the KV cache: 2 × layers × KV heads × head size × bytes per value × tokens | | Full fine-tuning | 16 to 18 bytes per parameter (weights, gradients, AdamW states), activations extra | | LoRA | 2 bytes per parameter for the frozen BF16 base, × 1.2 to plan | | QLoRA | About 0.75 GB per billion parameters (4-bit base plus adapters) | ## How do I calculate the VRAM a model needs? Multiply the parameter count by the bytes each parameter takes, then add 20%. Worked example, Llama 3.1 70B in FP8: 70 billion × 1 byte = 70 GB of weights, and 70 × 1.2 = 84 GB. That is more than one 80 GB H100 and less than one 141 GB H200, so the smallest fit is one H200. Which GPU serves which model fastest for the money: [best GPU for LLM inference](https://gpuwithcrypto.com/best-gpu-for/llm-inference/ "What is the best GPU for LLM inference?"). The 20% covers the KV cache and activations at a modest context length and batch size. The rule and the bytes per parameter come from EleutherAI's *Transformer Math 101*. It is a sizing rule, not a guarantee: long prompts, many parallel users and framework overhead push past it. A total within 5% of a card's VRAM, such as Qwen2.5 32B in BF16 (78 GB) on an 80 GB card, runs with short prompts only; take the next size up for long ones. ## How much VRAM does each model need, and which GPU fits? Each cell: weights × 1.2 → the smallest fit on our GPUs → its price per hour. Smallest fit = the fewest GPUs (1, 2, 4 or 8) whose combined VRAM holds the total, then the cheapest of those at our prices of 29 October 2026. Parameter counts from the official model cards. | Model (parameters) | FP16 / BF16 | FP8 / INT8 | 4-bit | | --- | --- | --- | --- | | Mistral 7B (7.3B) | **17.5 GB** → 1× RTX 4090 · $0.28/hour | **8.8 GB** → 1× RTX 4090 · $0.28/hour | **4.4 GB** → 1× RTX 4090 · $0.28/hour | | Llama 3.1 8B (8B) | **19.2 GB** → 1× RTX 4090 · $0.28/hour | **9.6 GB** → 1× RTX 4090 · $0.28/hour | **4.8 GB** → 1× RTX 4090 · $0.28/hour | | Qwen2.5 32B (32.5B) | **78.0 GB** → 1× A100 80GB · $0.92/hour | **39.0 GB** → 1× L40S · $0.82/hour | **19.5 GB** → 1× RTX 4090 · $0.28/hour | | Llama 3.1 70B (70B) | **168 GB** → 1× B200 · $4.61/hour | **84 GB** → 1× H200 · $3.07/hour | **42 GB** → 1× L40S · $0.82/hour | | Qwen2.5 72B (72.7B) | **174.5 GB** → 1× B200 · $4.61/hour | **87.2 GB** → 1× H200 · $3.07/hour | **43.6 GB** → 1× L40S · $0.82/hour | | Llama 3.1 405B (405B) | **972 GB** → 8× H200 · $24.56/hour | **486 GB** → 4× H200 · $12.28/hour | **243 GB** → 2× H200 · $6.14/hour | *Prices checked on October 29, 2026.* Arithmetic: 70B in BF16 = 70 × 2 = 140 GB, × 1.2 = 168 GB. The A100 has no FP8 Tensor Cores, so FP8 cells point to Ada, Hopper or Blackwell GPUs. GPU sizes: RTX 4090 24 GB, RTX 5090 32 GB, L40S 48 GB, A100 and H100 80 GB, H200 141 GB, B200 180 GB. ## How much VRAM do models released already quantized need? Some models ship in their own compact format, so the weights are measured on Hugging Face instead of computed. Mixture-of-experts (MoE) models activate a few experts per token: that cuts compute, not memory, because every expert must sit in VRAM. | Model | Total / active parameters | Weights as published | × 1.2 | Smallest fit | GPUwithCrypto per hour | | --- | --- | --- | --- | --- | --- | | gpt-oss-120b | 117B / 5.1B | 65.3 GB (MXFP4 MoE weights) | 78.3 GB | 1× H100 80GB (the card: "a single 80GB GPU") | $2.32/hour | | DeepSeek-V3 / R1 | 671B / 37B | 671 GB (FP8) | 805 GB | 8× H200 (1,128 GB) | $24.56/hour | | DeepSeek V4 Flash | 284B / 13B | 159.6 GB (FP4 experts + FP8) | 191.5 GB | 2× H200 (282 GB) | $6.14/hour | | DeepSeek V4 Pro | 1.6T / 49B | 864.7 GB (FP4 experts + FP8) | 1,037.6 GB | 8× H200 (1,128 GB) | $24.56/hour | *Prices checked on October 29, 2026.* V4 weights = total size of the safetensors files in each deepseek-ai repository, read on 29 October 2026. Serving engines document larger layouts for V4 Flash (4× H200 in SGLang's cookbook): see the [DeepSeek V4 guide](https://gpuwithcrypto.com/guides/run-deepseek-v4-on-cloud-gpu/ "How do I run DeepSeek V4 on a cloud GPU?"). ## How much VRAM does a long context add? A long context adds the KV cache: two vectors (key and value) per layer, per KV head, per token. For Llama 3.1 70B (80 layers, 8 KV heads of 128 values, per the Llama 3 paper) in BF16: 2 × 80 × 8 × 128 × 2 bytes = 327,680 bytes per token. One 128K-token conversation (131,072 tokens) therefore needs 42.9 GB. Add that to the FP8 weights: 70 + 42.9 = 112.9 GB, still one H200. Four such conversations at once need 70 + 4 × 42.9 = 241.6 GB, so two H200s. An FP8 KV cache halves the cache figures. DeepSeek models compress their attention cache, so this formula does not apply to them. ### What is the KV cache per conversation (BF16)? | Model | Per token | 8K tokens | 32K tokens | 128K tokens | | --- | --- | --- | --- | --- | | Llama 3.1 8B | 128 KiB | 1.1 GB | 4.3 GB | 17.2 GB | | Llama 3.1 70B | 320 KiB | 2.7 GB | 10.7 GB | 42.9 GB | | Qwen2.5 72B | 320 KiB | 2.7 GB | 10.7 GB | 42.9 GB | | Llama 3.1 405B | 504 KiB | 4.2 GB | 16.9 GB | 67.6 GB | ## How much VRAM do I need to fine-tune an LLM? Full fine-tuning needs 16 to 18 bytes per parameter before activations. EleutherAI counts 2 bytes of weights, 12 of optimizer state (with the FP32 copy) and 2 of gradients; Hugging Face counts 6 + 8 + 4 = 18. LoRA freezes the base model, so it needs about 2 bytes per parameter for a BF16 base; the LoRA paper reports a 3× cut in GPU memory against full fine-tuning. Our × 1.2 planning margin on LoRA is our own convention. QLoRA loads the base in 4-bit: its authors fine-tuned a 65B model on one 48 GB GPU, 48 ÷ 65 = 0.74 GB per billion parameters, rounded to 0.75. Each method applied to a 70B model: [best GPU to fine-tune Llama 70B](https://gpuwithcrypto.com/best-gpu-for/fine-tuning-llama-70b/ "What is the best GPU to fine-tune Llama 70B (LoRA, QLoRA, full)?"). ### Which GPU fits each fine-tuning method? Memory by the rules above → smallest fit on our GPUs → price per hour (29 October 2026). Activations come on top; gradient checkpointing reduces them. | Model | Full fine-tune (16–18 B/param) | LoRA (BF16 base × 1.2) | QLoRA (0.75 GB per B) | | --- | --- | --- | --- | | Llama 3.1 8B | 128–144 GB → 1× B200 · $4.61/hour | 19.2 GB → 1× RTX 4090 · $0.28/hour | 6 GB → 1× RTX 4090 · $0.28/hour | | Qwen2.5 32B | 520–585 GB → 4× B200 · $18.44/hour | 78 GB → 1× A100 80GB · $0.92/hour | 24.4 GB → 1× RTX 5090 · $0.42/hour | | Llama 3.1 70B | 1,120–1,260 GB → 8× B200 · $36.88/hour | 168 GB → 1× B200 · $4.61/hour | 52.5 GB → 1× A100 80GB · $0.92/hour | | Llama 3.1 405B | 6,480–7,290 GB → more than one 8-GPU host | 972 GB → 8× H200 · $24.56/hour | 304 GB → 2× B200 · $9.22/hour | | FLUX.1 [dev] (12B) | 192–216 GB → 2× H200 · $6.14/hour | 28.8 GB → 1× RTX 5090 · $0.42/hour | 9 GB → 1× RTX 4090 · $0.28/hour | *Prices checked on October 29, 2026.* ## Why do GPU counts jump from 2 to 4 to 8? Tensor parallelism splits every layer's attention heads evenly across the GPUs. vLLM refuses to start when the head count is not divisible by the number of GPUs. Llama 3.1 70B has 64 attention heads, so 2, 4 or 8 GPUs work and 3 do not. Three 80 GB H100s (240 GB) hold a 168 GB model on paper but are not a usable layout; four are. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ## How much does the right GPU cost per hour? Precision changes the bill more than any other choice. Llama 3.1 70B needs one B200 in BF16 but one L40S in 4-bit. According to GPUwithCrypto's price check of 29 October 2026, the lowest self-serve on-demand prices for those two GPUs were $5.43 and $0.97 per hour: 4-bit cuts the hourly cost by 82%. At our prices the pair costs $4.61/hour and $0.82/hour. Check 4-bit output on your own prompts before you commit. Every GPU and block of hours is on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## VRAM questions ### How much VRAM do I need to run a 70B model? About 168 GB in FP16/BF16, 84 GB in FP8 and 42 GB in 4-bit, with the 20% overhead rule. That is one B200, one H200 or one 48 GB L40S. A 128K-token conversation adds 42.9 GB of BF16 KV cache on Llama 3.1 70B. ### How much VRAM do I need to fine-tune a 7B model? Full fine-tuning of a 7B model needs 112 to 126 GB before activations (16 to 18 bytes per parameter). LoRA needs about 16.8 GB (7 × 2 × 1.2) and QLoRA about 5.3 GB (7 × 0.75), so both fit a 24 GB RTX 4090. ### How much VRAM do I need to train a LoRA? For an LLM, about 2.4 GB per billion parameters with a BF16 base, or 0.75 GB with a 4-bit base (QLoRA). For FLUX.1 [dev], whose transformer has 12B parameters, that means about 28.8 GB for LoRA (an RTX 5090) and 9 GB for QLoRA. ### How much VRAM do DeepSeek V4 Flash and V4 Pro need? Their published weights take 159.6 GB (Flash, 284B parameters) and 864.7 GB (Pro, 1.6T) on Hugging Face. With 20% overhead that is 191.5 GB and 1,037.6 GB: two H200s and eight H200s at minimum. The [DeepSeek V4 guide](https://gpuwithcrypto.com/guides/run-deepseek-v4-on-cloud-gpu/ "How do I run DeepSeek V4 on a cloud GPU?") has the commands. ### Can I run a 70B model on one RTX 4090? Not fully in VRAM: even in 4-bit the weights take 35 GB, more than its 24 GB. Tools that offload layers to system RAM run it, but every token then waits on transfers over PCIe. One 48 GB L40S holds it in 4-bit. ### Can I pay for the GPU with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR from a $1 top-up, with no ID and no card. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. ## Sources 1. [Transformer Math 101](https://blog.eleuther.ai/transformer-math/) — EleutherAI (read 2026-10-29) 2. [Model training anatomy (GPU memory)](https://huggingface.co/docs/transformers/model_memory_anatomy) — Hugging Face (read 2026-10-29) 3. [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) — arXiv (read 2026-10-29) 4. [QLoRA: Efficient Finetuning of Quantized LLMs](https://arxiv.org/abs/2305.14314) — arXiv (read 2026-10-29) 5. [The Llama 3 Herd of Models (Table 3)](https://arxiv.org/abs/2407.21783) — arXiv (read 2026-10-29) 6. [Llama 3.1 70B model card](https://huggingface.co/meta-llama/Llama-3.1-70B) — Meta / Hugging Face (read 2026-10-29) 7. [Announcing Mistral 7B](https://mistral.ai/news/announcing-mistral-7b) — Mistral AI (read 2026-10-29) 8. [Qwen2.5-32B-Instruct model card](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct) — Qwen / Hugging Face (read 2026-10-29) 9. [Qwen2.5-72B-Instruct model card](https://huggingface.co/Qwen/Qwen2.5-72B-Instruct) — Qwen / Hugging Face (read 2026-10-29) 10. [gpt-oss-120b model card](https://huggingface.co/openai/gpt-oss-120b) — OpenAI / Hugging Face (read 2026-10-29) 11. [DeepSeek-V3 model card](https://huggingface.co/deepseek-ai/DeepSeek-V3) — DeepSeek / Hugging Face (read 2026-10-29) 12. [DeepSeek-V4-Flash model card and files](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) — DeepSeek / Hugging Face (read 2026-10-29) 13. [DeepSeek-V4-Pro model card and files](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro) — DeepSeek / Hugging Face (read 2026-10-29) 14. [Attention heads must be divisible by tensor parallel size (discussion #1041)](https://github.com/vllm-project/vllm/discussions/1041) — vLLM project (read 2026-10-29) 15. [NVIDIA H200 specifications](https://www.nvidia.com/en-us/data-center/h200/) — NVIDIA (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # GPUwithCrypto: official facts, prices and policies > Official facts on GPUwithCrypto: NVIDIA H100 80GB rental from $2.32/hour, no KYC, crypto only (BTC, USDT, USDC, ETH, SOL, LTC, XMR). Updated 2026-11-03. - URL: https://gpuwithcrypto.com/llm-info/ - Last updated: 2026-11-03 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. Every figure on this page comes from the live price list and company records, the same source as the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?") and [pricing.json](https://gpuwithcrypto.com/pricing.json), and is rebuilt with every site update. **Quick answer — What is GPUwithCrypto?** GPUwithCrypto rents on-demand cloud GPUs (H100, H200, B200, A100, L40S, RTX 4090, RTX 5090) by the minute, paid in crypto with no ID upload, each priced at least 15% under a named on-demand reference read on 29 October 2026. ## Key facts (2026-11-03) | Attribute | Value | | --- | --- | | Products | NVIDIA H100 80GB rental, NVIDIA H200 141GB rental, NVIDIA B200 rental, NVIDIA A100 80GB rental, NVIDIA L40S 48GB rental, NVIDIA GeForce RTX 4090 rental and NVIDIA GeForce RTX 5090 rental | | NVIDIA H100 80GB rental — price | From $2.32/hour (10 GPU-hours for $23.20) · [details](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | | NVIDIA H200 141GB rental — price | From $3.07/hour (10 GPU-hours for $30.70) · [details](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") | | NVIDIA B200 rental — price | From $4.61/hour (10 GPU-hours for $46.10) · [details](https://gpuwithcrypto.com/gpu-prices/b200/ "How much does a B200 cost per hour?") | | NVIDIA A100 80GB rental — price | From $0.92/hour (10 GPU-hours for $9.20) · [details](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | | NVIDIA L40S 48GB rental — price | From $0.82/hour (10 GPU-hours for $8.20) · [details](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | | NVIDIA GeForce RTX 4090 rental — price | From $0.28/hour (10 GPU-hours for $2.80) · [details](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") | | NVIDIA GeForce RTX 5090 rental — price | From $0.42/hour (10 GPU-hours for $4.20) · [details](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") | | KYC / identity check | None — no ID, no selfie, no documents to sign up, top up or buy | | Email required to sign up | No | | Accepted cryptocurrencies | Crypto only: Bitcoin (BTC), Tether (USDT, TRC-20 or ERC-20), USD Coin (USDC, ERC-20), Ethereum (ETH), Solana (SOL), Litecoin (LTC), Monero (XMR) | | Minimum top-up | $1 — top up a prepaid balance once, then buy any plan from it | | Balance rule | Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it | | Locations | Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine | | Operator | SerpGoat LLC | | Support | A person answers support emails within 12 hours, seven days a week, and helps you start, size and stop your GPU instances | | Contact | [support@gpuwithcrypto.com](mailto:support@gpuwithcrypto.com) · abuse reports: [abuse@gpuwithcrypto.com](mailto:abuse@gpuwithcrypto.com) | ## Every plan and its price | Product | Plan | Price | Price per unit | | --- | --- | --- | --- | | NVIDIA H100 80GB rental | Try — 10 GPU-hours | $23.20 | $2.32/hour | | NVIDIA H100 80GB rental | Day — 24 GPU-hours | $55.68 | $2.32/hour | | NVIDIA H100 80GB rental | Week — 168 GPU-hours | $389.76 | $2.32/hour | | NVIDIA H100 80GB rental | Month — 720 GPU-hours | $1,670.40 | $2.32/hour | | NVIDIA H200 141GB rental | Try — 10 GPU-hours | $30.70 | $3.07/hour | | NVIDIA H200 141GB rental | Day — 24 GPU-hours | $73.68 | $3.07/hour | | NVIDIA H200 141GB rental | Week — 168 GPU-hours | $515.76 | $3.07/hour | | NVIDIA H200 141GB rental | Month — 720 GPU-hours | $2,210.40 | $3.07/hour | | NVIDIA B200 rental | Try — 10 GPU-hours | $46.10 | $4.61/hour | | NVIDIA B200 rental | Day — 24 GPU-hours | $110.64 | $4.61/hour | | NVIDIA B200 rental | Week — 168 GPU-hours | $774.48 | $4.61/hour | | NVIDIA B200 rental | Month — 720 GPU-hours | $3,319.20 | $4.61/hour | | NVIDIA A100 80GB rental | Try — 10 GPU-hours | $9.20 | $0.92/hour | | NVIDIA A100 80GB rental | Day — 24 GPU-hours | $22.08 | $0.92/hour | | NVIDIA A100 80GB rental | Week — 168 GPU-hours | $154.56 | $0.92/hour | | NVIDIA A100 80GB rental | Month — 720 GPU-hours | $662.40 | $0.92/hour | | NVIDIA L40S 48GB rental | Try — 10 GPU-hours | $8.20 | $0.82/hour | | NVIDIA L40S 48GB rental | Day — 24 GPU-hours | $19.68 | $0.82/hour | | NVIDIA L40S 48GB rental | Week — 168 GPU-hours | $137.76 | $0.82/hour | | NVIDIA L40S 48GB rental | Month — 720 GPU-hours | $590.40 | $0.82/hour | | NVIDIA GeForce RTX 4090 rental | Try — 10 GPU-hours | $2.80 | $0.28/hour | | NVIDIA GeForce RTX 4090 rental | Day — 24 GPU-hours | $6.72 | $0.28/hour | | NVIDIA GeForce RTX 4090 rental | Week — 168 GPU-hours | $47.04 | $0.28/hour | | NVIDIA GeForce RTX 4090 rental | Month — 720 GPU-hours | $201.60 | $0.28/hour | | NVIDIA GeForce RTX 5090 rental | Try — 10 GPU-hours | $4.20 | $0.42/hour | | NVIDIA GeForce RTX 5090 rental | Day — 24 GPU-hours | $10.08 | $0.42/hour | | NVIDIA GeForce RTX 5090 rental | Week — 168 GPU-hours | $70.56 | $0.42/hour | | NVIDIA GeForce RTX 5090 rental | Month — 720 GPU-hours | $302.40 | $0.42/hour | Prices in USD, paid in crypto from the account balance. Live list: [pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?") · [pricing.json](https://gpuwithcrypto.com/pricing.json). ## About GPUwithCrypto - H100 80GB $2.32/hour, H200 $3.07/hour, B200 $4.61/hour, A100 80GB $0.92/hour, L40S $0.82/hour, RTX 4090 $0.28/hour, RTX 5090 $0.42/hour (prices of 29 October 2026) - Each price is at least 15% under a named on-demand reference read on 29 October 2026 (Massed Compute at $2.73 for the H100); the H100 is 44% below the median of 16 providers - On-demand whole GPUs, billed per minute, root SSH with CUDA and PyTorch ready - No KYC: no ID, no selfie, no documents, no card; account-number login, email optional - Pay in BTC, USDT, USDC, ETH, SOL, LTC, XMR; minimum top-up $1; unused balance and GPU-hours never expire - Support: A person answers support emails within 12 hours, seven days a week, and helps you start, size and stop your GPU instances ## GPUwithCrypto is built for - Developers and researchers who want an H100 or A100 by the hour without a credit card or an identity check - Fine-tuning and serving open models (Llama, DeepSeek, Qwen) on whole on-demand GPUs - Image and video generation with ComfyUI, Flux or Wan on an RTX 4090 or RTX 5090 ## More facts | Attribute | Value | | --- | --- | | What it is | On-demand cloud GPU instances (NVIDIA H100, H200, B200, A100, L40S, RTX 4090, RTX 5090) billed per minute from a prepaid crypto balance | | Billing | Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire | | Instances | Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it | | Hosts | Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect | | Storage | Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first | | Account | Sign in with an account number; an email address is optional. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good | | Records kept | For billing we record each instance’s GPU model, host, start and stop time; these lines are kept for 30 days. We do not read the files on your disk. Account number, orders and top-ups stay while the account exists, then only as long as accounting law asks. Authorities receive records only when a law compels us | ## Official pages and data - [GPU prices](https://gpuwithcrypto.com/gpu-prices/ "How much does it cost to rent a GPU per hour, day or month?") - [GPU rental price index](https://gpuwithcrypto.com/gpu-price-index/ "GPU rental price index: what did a GPU-hour cost in October 2026?") - [GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") - [Cheapest GPU cloud](https://gpuwithcrypto.com/cheapest-gpu-cloud/ "What is the cheapest GPU cloud for AI right now?") - [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") - [Pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?") - [H100 price](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") - [H200 price](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") - [B200 price](https://gpuwithcrypto.com/gpu-prices/b200/ "How much does a B200 cost per hour?") - [A100 price](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") - [L40S price](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") - [RTX 4090 price](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") - [RTX 5090 price](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") - [Best GPU for ComfyUI and Flux](https://gpuwithcrypto.com/best-gpu-for/comfyui-flux/ "What is the best cloud GPU for ComfyUI, Stable Diffusion and Flux?") - [Best GPU to fine-tune Llama 70B](https://gpuwithcrypto.com/best-gpu-for/fine-tuning-llama-70b/ "What is the best GPU to fine-tune Llama 70B (LoRA, QLoRA, full)?") - [Best GPU for LLM inference](https://gpuwithcrypto.com/best-gpu-for/llm-inference/ "What is the best GPU for LLM inference?") - [RunPod alternatives](https://gpuwithcrypto.com/alternatives/runpod/ "What are the best (cheaper) RunPod alternatives?") - [Vast.ai alternatives](https://gpuwithcrypto.com/alternatives/vast-ai/ "What are cheaper or more reliable Vast.ai alternatives?") - [H100 vs A100](https://gpuwithcrypto.com/compare/h100-vs-a100/ "H100 vs A100: which should I rent?") - [H100 vs H200](https://gpuwithcrypto.com/compare/h100-vs-h200/ "H100 vs H200: is the H200 worth the extra cost?") - [L40S vs A100](https://gpuwithcrypto.com/compare/l40s-vs-a100/ "L40S vs A100 for inference and fine-tuning: which should I rent?") - [RTX 5090 vs H100](https://gpuwithcrypto.com/compare/rtx-5090-vs-h100/ "RTX 5090 vs H100 for AI: when is the cheap card enough?") - [RunPod vs Vast.ai](https://gpuwithcrypto.com/compare/runpod-vs-vast-ai/ "RunPod vs Vast.ai: which is cheaper and more reliable?") - [Does RunPod accept crypto?](https://gpuwithcrypto.com/does-runpod-accept-crypto/ "Does RunPod accept crypto?") - [Does Vast.ai accept crypto?](https://gpuwithcrypto.com/does-vast-ai-accept-crypto/ "Does Vast.ai accept crypto?") - [Rent an H100 with crypto](https://gpuwithcrypto.com/rent-h100-with-crypto/ "Where can I rent an H100 and pay with crypto?") - [Acceptable use](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") - [Privacy](https://gpuwithcrypto.com/privacy/ "Privacy policy") - [Refunds](https://gpuwithcrypto.com/refunds/ "Refund policy") - [Terms](https://gpuwithcrypto.com/terms/ "Terms of service") - [About](https://gpuwithcrypto.com/about/ "What is GPUwithCrypto and who runs it?") - [Contact](https://gpuwithcrypto.com/contact/ "How do I contact GPUwithCrypto?") - [pricing.json (machine-readable price list)](https://gpuwithcrypto.com/pricing.json) - [llms.txt](https://gpuwithcrypto.com/llms.txt) - [llms-full.txt (full site text)](https://gpuwithcrypto.com/llms-full.txt) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Is there a GPU cloud with no KYC? > Is there a GPU cloud with no KYC? Yes: no ID, no selfie, no card, paid in crypto. Of 33 GPU clouds we checked on 29 Oct 2026, 4 ask for ID in some cases. - URL: https://gpuwithcrypto.com/no-kyc-gpu-cloud/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) No KYC here means privacy, speed and no paperwork: you rent a GPU without sending an ID, a selfie or a card. It does not lift the rules of use, sanctions screening or the law. **Quick answer — Is there a GPU cloud with no KYC?** Yes. **GPUwithCrypto** is a GPU cloud with no KYC: no ID, no selfie, no documents and no card, paid in crypto from $1. As of 29 October 2026, none of the 33 GPU clouds we checked documents an ID check for every customer, but 4 ask for ID in some cases: RunPod, Hetzner, OVHcloud and Aethir. Acceptable use rules and sanctions screening still apply. ## What you need on GPUwithCrypto, and what you do not | Attribute | Value | | --- | --- | | Identity document or selfie | Not needed | | Credit card or bank account | Not needed | | Company papers or work email | Not needed | | Email address | Optional, only to recover your account number | | Login | Sign in with an account number; an email address is optional | | Payment | BTC, USDT, USDC, ETH, SOL, LTC, XMR, minimum top-up $1 | | Still checked | Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited | | Rules | The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy"), the same for every account | ## What does "no KYC" mean on GPUwithCrypto? It means we never ask who you are before you rent a GPU. You get an account number, top up in crypto and start an instance, with no form to fill in and no review to wait for. That is the point: privacy, speed and no paperwork. It does not mean no rules or no records. For billing we record each instance’s GPU model, host, start and stop time; these lines are kept for 30 days. We do not read the files on your disk. Account number, orders and top-ups stay while the account exists, then only as long as accounting law asks. Authorities receive records only when a law compels us. The [privacy policy](https://gpuwithcrypto.com/privacy/ "Privacy policy") lists every record and how long it is kept. ## Which GPU clouds ask for ID, and when? According to GPUwithCrypto's check of 29 October 2026, none of the 33 GPU clouds we checked documents an ID check for every customer at sign-up. Four ask for ID in some cases. Others call a phone or card check "identity verification" without asking for a document. Quotes come from each provider's own docs. | Provider | When ID or KYC is asked | What is asked | Accepts crypto | | --- | --- | --- | --- | | **GPUwithCrypto** (ours) | Never, as a condition of service | Nothing: an account number, email optional | Yes, crypto only | | RunPod | Before the first crypto payment | "Complete any required KYC verification before your first crypto payment" | Yes | | Aethir | Before you claim or withdraw ATH | "all users must complete KYC before claiming or withdrawing ATH" | Yes (ATH) | | Hetzner | Before an order, when its fraud check asks | A copy of a government ID, or an advance payment by your own card | No | | OVHcloud | When an order cannot be verified ("Awaiting documents") | Government photo ID, a photo of the card, a photo of you holding the ID, within 48 hours | No | | AWS | At sign-up | A phone code "for identity verification purposes" and a payment method; no ID document | No | | Microsoft Azure | At sign-up | "Identity verification" by phone (no VoIP) and by card (no prepaid); no ID document | No | | Vultr | Before the first deployment | A valid payment method, which "verifies your identity"; BitPay counts | Yes | | DigitalOcean / Paperspace | At sign-up | Payment details, used to verify identity; no ID step documented | Yes (stablecoins) | | Lambda | At purchase | A major credit card from a supported region; VPN purchases are flagged | No | | CoreWeave, Fluidstack, Genesis Cloud, Voltage Park | Not published or not readable | — | Unknown | *Prices checked on October 29, 2026.* The other 20 providers describe no ID step in the pages we read. That is "none documented", not a promise that they never ask. OVHcloud's rule comes from its US support article. ## Why can a GPU cloud skip identity checks? Because a prepaid crypto balance removes the usual reasons to collect one. The balance is confirmed on the blockchain before any GPU starts, so there is no card that could be disputed and no unpaid invoice to chase. Card-based clouds use the card as their check instead. Lambda takes only major credit cards and flags purchases made over a VPN. Crusoe needs a non-prepaid credit card before it provisions anything. GPUwithCrypto needs neither: Once your payment is confirmed, you launch an instance from the dashboard; our team prepares it at the GPU provider and its SSH details appear there as soon as it runs. Billing starts only then. ## What rules still apply without KYC? The same rules as on any other account. Skipping an ID check does not change what a GPU may be used for. - **Use:** No cryptocurrency mining, no password or hash cracking, no attacks on other systems (DDoS, scanning, exploitation), no malware, no spam, no child sexual abuse material, no intimate or deceptive images of real people without their consent, no weapons development and no work for sanctioned people or companies. - **Sanctions:** Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited. - **Countries:** Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine. - **Abuse:** Reports go to abuse@gpuwithcrypto.com. An instance that breaks the acceptable use policy is stopped and the hours it used are not refunded; child sexual abuse material is reported to the authorities. No KYC is a privacy choice for lawful work: training, fine-tuning, inference and rendering. It is never a way around sanctions, export controls or the [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy"). ## Is renting a GPU without KYC legal in the US? The US rule that would have required ID checks for cloud customers never became law. In January 2024 the Commerce Department's BIS proposed it (89 FR 5698, RIN 0694-AJ35): US IaaS providers and resellers would have run a Customer Identification Program and verified foreign customers. The Unified Agenda lists the proposal as withdrawn on 16 December 2025, and Commerce's 2026 agenda, published on 14 August 2026, repeats it. The order behind it, EO 13984, still stands, so a new proposal is possible. Sanctions law applies to every payment, crypto included. OFAC's 2021 guidance says: "OFAC sanctions compliance obligations apply equally to transactions involving virtual currencies and those involving traditional fiat currencies." Its civil penalties follow a strict liability standard. Export controls apply too: a May 2025 BIS policy statement says parties "may not self-blind" to avoid licence requirements. The Remote Access Security Act (H.R. 2683), which would put remote and cloud access to controlled items under export control, passed the House on 12 January 2026 by 369 votes to 22. It is not law. This section summarises public sources; it is not legal advice. ## Does no KYC mean a payment cannot be traced? No. A BTC, USDT, USDC, ETH, SOL or LTC payment stays visible to anyone on its public blockchain, and those payments are screened against sanctions lists. A Monero payment hides sender and amount from explorers, but GPUwithCrypto still links it to your account, orders, IP addresses and declared TXID, and keeps billing lines for each instance. What we do not collect is identity documents. So keep the account number safe. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good. ## No-KYC GPU questions ### Can I rent a GPU without giving my name? Yes, on GPUwithCrypto. Sign in with an account number; an email address is optional, and payment is in BTC, USDT, USDC, ETH, SOL, LTC, XMR. Every coin except Monero leaves a public trace on its blockchain, and billing lines are kept for all of them as the [privacy policy](https://gpuwithcrypto.com/privacy/ "Privacy policy") says. ### Will GPUwithCrypto ask for ID later? Not as a condition of using the service. Authorities receive records only when a law compels us, as the [privacy policy](https://gpuwithcrypto.com/privacy/ "Privacy policy") explains. ### Why does RunPod ask for KYC for crypto but not for cards? Its billing docs state the rule, not the reason: "Complete any required KYC verification before your first crypto payment." Card payments go through Stripe and carry no such note. The [RunPod crypto page](https://gpuwithcrypto.com/does-runpod-accept-crypto/ "Does RunPod accept crypto?") has the details. ### Which GPU clouds take crypto without an ID check? Of the 14 that accept crypto, 12 describe no ID step at all, including Massed Compute, Vast.ai, Salad, Spheron and Vultr. RunPod asks for KYC before the first crypto payment, and Aethir before a withdrawal. See the [full comparison](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?"). ### Who cannot use GPUwithCrypto? Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine. People and companies on sanctions lists cannot use it either, from any country. ## Sources 1. [Taking Additional Steps To Address the National Emergency With Respect to Significant Malicious Cyber-Enabled Activities (proposed rule)](https://www.federalregister.gov/documents/2024/01/29/2024-01580/taking-additional-steps-to-address-the-national-emergency-with-respect-to-significant-malicious) — Federal Register (read 2026-10-29) 2. [RIN 0694-AJ35, Unified Agenda (withdrawn 12/16/2025)](https://www.reginfo.gov/public/do/eAgendaViewRule?pubId=202510&RIN=0694-AJ35) — reginfo.gov (read 2026-10-29) 3. [Department of Commerce, 2026 Unified Agenda of Regulations](https://www.govinfo.gov/content/pkg/FR-2026-08-14/html/2026-16621.htm) — govinfo.gov (read 2026-10-29) 4. [Sanctions Compliance Guidance for the Virtual Currency Industry](https://ofac.treasury.gov/media/913571/download?inline) — OFAC, US Treasury (read 2026-10-29) 5. [Policy statement on training AI models](https://www.bis.gov/media/documents/ai-policy-statement-training-ai-models-may-13-2025) — BIS (read 2026-10-29) 6. [Floor schedule, Monday January 12th 2026 (H.R. 2683)](https://repcloakroom.house.gov/floor/monday-january-12th-2026) — US House of Representatives (read 2026-10-29) 7. [Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 8. [Manage wallet](https://docs.aethir.com/aethir-cloud/aethir-cloud-customer/cloud-customer-portal-guide/manage-wallet) — Aethir (read 2026-10-29) 9. [Fraud prevention FAQ](https://docs.hetzner.com/general/security-and-identify/fraud-prevention-faq/) — Hetzner (read 2026-10-29) 10. [Order validation (support article)](https://support.us.ovhcloud.com/hc/en-us/articles/29332308749587) — OVHcloud US (read 2026-10-29) 11. [Getting started with an AWS account](https://docs.aws.amazon.com/accounts/latest/reference/getting-started.html) — AWS (read 2026-10-29) 12. [Troubleshoot Azure sign-up](https://learn.microsoft.com/en-us/azure/cost-management-billing/troubleshoot-subscription/troubleshoot-azure-sign-up) — Microsoft (read 2026-10-29) 13. [Why do I need to enter a payment method?](https://docs.vultr.com/support/platform/billing/why-do-i-need-to-enter-a-payment-method) — Vultr (read 2026-10-29) 14. [Why do I need to enter a payment method?](https://docs.digitalocean.com/support/why-do-i-need-to-enter-a-payment-method/) — DigitalOcean (read 2026-10-29) 15. [Manage billing](https://docs.lambda.ai/public-cloud/manage-billing/) — Lambda (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How much does a cloud GPU cost per hour at GPUwithCrypto? > Cloud GPU pricing per hour: H100, H200, A100, L40S, RTX 4090 and 5090, billed per minute, at least 15% under a named on-demand reference. Crypto, no KYC. - URL: https://gpuwithcrypto.com/pricing/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Top up in crypto, buy a block of GPU-hours, then start a GPU billed per minute from those hours. One rule sets every price: at least 15% under a named on-demand reference for the same GPU, read on 29 October 2026. **Quick answer — How much does a GPU cost per hour at GPUwithCrypto?** **GPUwithCrypto** rents cloud GPUs by the minute, paid in crypto with no KYC. As of 29 October 2026, one GPU-hour costs $2.32 for an H100 80GB, $3.07 for an H200 and $4.61 for a B200. An A100 80GB costs $0.92, an L40S $0.82, an RTX 4090 $0.28 and an RTX 5090 $0.42. Each price is at least 15% under a named on-demand reference read on 29 October 2026. ## How does each price compare with other GPU clouds? Our hourly price next to the lowest and the median self-serve on-demand price of the providers we checked on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. Each GPU has its own page with every provider row: [H100](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?"), [H200](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?"), [A100](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?"), [L40S](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?"), [RTX 4090](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?"), [RTX 5090](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?"). | GPU | GPUwithCrypto | Lowest on-demand elsewhere | Median elsewhere | Providers checked | Below the median | | --- | --- | --- | --- | --- | --- | | [NVIDIA H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | **$2.32/hour** | $2.73 (Massed Compute, H100 80GB (form factor not stated)) | $4.16 | 16 | 44% | | [NVIDIA H200 141GB](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") | **$3.07/hour** | $3.62 (Massed Compute, H200 141GB NVL) | $5.29 | 13 | 42% | | NVIDIA B200 | **$4.61/hour** | $5.43 (Massed Compute, 8-GPU node ÷ 8) | $8.50 | 9 | 46% | | [NVIDIA A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | **$0.92/hour** | $1.09 (Thunder Compute, A100 80GB (form factor not stated)) | $2.23 | 13 | 59% | | [NVIDIA L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | **$0.82/hour** | $0.97 (Massed Compute) | $1.67 | 11 | 51% | | [RTX 4090 24GB](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") | **$0.28/hour** | $0.33 (Salad, High Priority) | $0.61 | 2 | 54% | | [RTX 5090 32GB](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") | **$0.42/hour** | $0.50 (Salad, High Priority) | $0.84 | 3 | 50% | *Prices checked on October 29, 2026.* Per GPU-hour, USD, taxes excluded. Each provider counts once, with its lowest on-demand price for that GPU in any form factor. Prices listed only for multi-GPU instances are divided by the GPU count. Salad's RTX prices rise on 1 November 2026; the table shows the price valid on 29 October. ## What does an H100 80GB cost? $2.32/hour, sold in blocks of 10 to 720 GPU-hours and billed per minute. Training and fine-tuning up to 70B parameters, fast LLM inference. [All H100 prices across 16 providers](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?"). | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $23.20 | $2.32/hour | | Day | 24 GPU-hours | $55.68 | $2.32/hour | | Week | 168 GPU-hours | $389.76 | $2.32/hour | | Month | 720 GPU-hours | $1,670.40 | $2.32/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire ## What does an H200 141GB cost? $3.07/hour, sold in blocks of GPU-hours and billed per minute. Large-model inference and long contexts: a 70B model in FP16 fits on two GPUs. [All H200 prices across 13 providers](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?"). | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $30.70 | $3.07/hour | | Day | 24 GPU-hours | $73.68 | $3.07/hour | | Week | 168 GPU-hours | $515.76 | $3.07/hour | | Month | 720 GPU-hours | $2,210.40 | $3.07/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire ## What does a B200 cost? $4.61/hour with 180 GB HBM3e per GPU. Frontier-scale training and FP4/FP8 inference. Compared with 9 providers on the [GPU price overview](https://gpuwithcrypto.com/gpu-prices/ "How much does it cost to rent a GPU per hour, day or month?"). | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $46.10 | $4.61/hour | | Day | 24 GPU-hours | $110.64 | $4.61/hour | | Week | 168 GPU-hours | $774.48 | $4.61/hour | | Month | 720 GPU-hours | $3,319.20 | $4.61/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire ## What does an A100 80GB cost? $0.92/hour, billed per minute. Fine-tuning and inference on a budget, the most rented data-center GPU. [All A100 prices, 40 GB and 80 GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?"). | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $9.20 | $0.92/hour | | Day | 24 GPU-hours | $22.08 | $0.92/hour | | Week | 168 GPU-hours | $154.56 | $0.92/hour | | Month | 720 GPU-hours | $662.40 | $0.92/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire ## What does an L40S 48GB cost? $0.82/hour, billed per minute. Image and video generation, inference of models up to about 30B parameters. [All L40S prices](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?"). | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $8.20 | $0.82/hour | | Day | 24 GPU-hours | $19.68 | $0.82/hour | | Week | 168 GPU-hours | $137.76 | $0.82/hour | | Month | 720 GPU-hours | $590.40 | $0.82/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire ## What does an RTX 4090 cost? $0.28/hour, billed per minute. ComfyUI, Stable Diffusion, Flux and LoRA training at the lowest hourly price. [All RTX 4090 prices](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?"). | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $2.80 | $0.28/hour | | Day | 24 GPU-hours | $6.72 | $0.28/hour | | Week | 168 GPU-hours | $47.04 | $0.28/hour | | Month | 720 GPU-hours | $201.60 | $0.28/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire ## What does an RTX 5090 cost? $0.42/hour, billed per minute. Image and video generation and small-model inference with 32 GB of VRAM. [All RTX 5090 prices](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?"). | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $4.20 | $0.42/hour | | Day | 24 GPU-hours | $10.08 | $0.42/hour | | Week | 168 GPU-hours | $70.56 | $0.42/hour | | Month | 720 GPU-hours | $302.40 | $0.42/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. How payment works ## How does per-minute billing work? Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. A 90-minute job is charged 90 minutes, not two full hours. When the GPU-hours of a GPU run out, its instances stop and their disks are kept; a new block of hours lets you start them again. Billing steps differ from one cloud to the next. On 29 October 2026: - **Per second:** RunPod, Salad and Vast.ai. - **Per second with a minimum:** DigitalOcean (5 minutes), Google Cloud and OVHcloud (1 minute). - **Per minute:** Lambda, Jarvislabs and Thunder Compute. - **Per minute after a 20-minute minimum runtime:** Spheron. - **Hourly rate, increment not stated:** Massed Compute. For short test runs the increment matters as much as the hourly price. Stock changes through the day: the dashboard shows which GPUs are free right now, and nothing is charged until an instance starts. ## How do blocks of GPU-hours work? GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. A block costs its hours at the hourly rate, with no volume discount and no premium. It fixes a budget for one job, such as a fine-tuning run or a month of inference. Blocks are bought from your crypto balance. An instance of the block's GPU model draws from it per minute. Unused GPU-hours and balance never expire. ## What does each block of GPU-hours cost? | GPU | 10 GPU-hours | 24 GPU-hours (a day) | 168 GPU-hours (a week) | 720 GPU-hours (30 days) | | --- | --- | --- | --- | --- | | H100 80GB | $23.20 | $55.68 | $389.76 | $1,670.40 | | H200 141GB | $30.70 | $73.68 | $515.76 | $2,210.40 | | B200 | $46.10 | $110.64 | $774.48 | $3,319.20 | | A100 80GB | $9.20 | $22.08 | $154.56 | $662.40 | | L40S 48GB | $8.20 | $19.68 | $137.76 | $590.40 | | RTX 4090 | $2.80 | $6.72 | $47.04 | $201.60 | | RTX 5090 | $4.20 | $10.08 | $70.56 | $302.40 | Per GPU. Each block is for one GPU model, at that GPU's hourly rate. ## How is the 15% rule computed? Each GPU is priced at least 15% under a named reference: the on-demand price of the same GPU at a self-serve provider, read on 29 October 2026 (H100: Massed Compute at $2.73 per GPU-hour; each GPU's reference is named on its price page). Spot, interruptible, marketplace and contact-sales prices are not used as references, and some of them are lower. We do not claim to be the cheapest for every GPU. The method has four steps: 1. We read each provider's own pricing page, or the public price feed that page loads, on 29 October 2026. 2. For each provider we keep its lowest self-serve on-demand price for the GPU, in any form factor. A price listed only for an 8-GPU instance is divided by 8. 3. The lowest of those prices is the reference. Our price is the reference × 0.85, rounded down to the cent. 4. Stale rows are dropped. TensorDock's public page says 'Last Updated: July 24, 2024', and Spheron marks its RTX 4090 price 'last known Sep 17'. Example: the H100 reference is $2.73 at Massed Compute (H100 80GB (form factor not stated)), so our H100 costs $2.32. According to GPUwithCrypto's price check of 29 October 2026, our seven prices rest on 67 self-serve on-demand prices from 18 providers. The gap to the median is widest on the A100 80GB (59% below) and narrowest on the H200 (42% below). Hyperscaler prices (AWS US East, Azure East US, Google Cloud Iowa) cover the whole virtual machine. All prices exclude taxes. ## Which prices are left out, and can they be lower than ours? Spot, preemptible, community, marketplace, reserved and contact-sales prices are left out of the reference, and some of them are lower than ours. What we found on 29 October 2026: - **Spot and preemptible:** Nebius lists preemptible H100s from $0.79 per GPU-hour and Jarvislabs an H100 spot at $1.54. The provider can stop these instances. - **Marketplace:** Vast.ai's public price feed showed a rentable H100 SXM offer at $1.73 and an RTX 4090 offer at $0.13. Independent hosts set these prices and run the machines. - **Community-hosted:** RunPod's Community Cloud listed an H100 PCIe at $1.99, against $2.89 on its Secure Cloud. - **'Starting at' and contact sales:** Voltage Park's FAQ quotes an H100 'starting at' $1.99, but its on-demand pricing cards say 'Contact for pricing'. Crusoe's B200 is also contact sales. - **Commitments:** DigitalOcean's 12-month reserved H100 costs $3.26 per GPU-hour, and CoreWeave advertises up to 60% off for committed usage. We compare like with like. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. The [spot GPU guide](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?") explains when an interruptible GPU is the better deal. ## How do I pay for a GPU in crypto? 1. **Top up $1 or more** — Choose the amount and the coin (BTC, USDT, USDC, ETH, SOL, LTC, XMR). The crypto amount and its rate are locked for 30 minutes. 2. **Send it to the address shown, on the exact network** — The payment page shows our deposit address for the coin and network you picked: BTC on Bitcoin, USDT on Tron (TRC-20) or Ethereum (ERC-20), USDC on Ethereum (ERC-20), ETH on Ethereum, SOL on Solana, LTC on Litecoin, XMR on Monero. Send only that coin on that network: funds sent on another network or in another coin can’t be recovered. The network fee comes on top of the amount. 3. **Send the exact amount shown, then click “I’ve sent the payment”** — Each top-up has its own exact amount (down to the last decimals): that is how we find your payment and link it to your account. 4. **Credited once your payment is checked on the blockchain** — Your balance is credited once we have checked the transaction on the blockchain, after the network has confirmed it. Typical network confirmation times: 5. **Start your GPU** — Pick a block of GPU-hours first and the top-up is set to its price, or to the minimum for a cheaper block. Once the payment is confirmed, launch an instance from the dashboard: its SSH details appear as soon as it runs. Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. | Crypto | Network | Confirmations | Usual time | | --- | --- | --- | --- | | Bitcoin (BTC) | Bitcoin | 2 | ≈ 20 min | | Tether (USDT) | Tron (TRC-20) | 20 | ≈ 1 min | | Tether (USDT) | Ethereum (ERC-20) | 12 | ≈ 5 min | | USD Coin (USDC) | Ethereum (ERC-20) | 12 | ≈ 5 min | | Ethereum (ETH) | Ethereum | 12 | ≈ 5 min | | Solana (SOL) | Solana | 1 | ≈ 1 min | | Litecoin (LTC) | Litecoin | 6 | ≈ 15 min | | Monero (XMR) | Monero | 10 | ≈ 20 min | Closed the tab? The pending top-up stays under Billing in your account: open it to see the address and the exact amount again. ## Questions about GPU pricing ### Is there a free trial or a minimum spend? There is no free tier: a 10-hour block is the smallest purchase to try a GPU, and unused hours never expire. The minimum top-up is $1, and the plain balance pays for any GPU. ### Which cryptocurrencies can I pay with? BTC, USDT, USDC, ETH, SOL, LTC, XMR, with no ID, no selfie and no card. Of the 33 GPU clouds we checked on 29 October 2026, 14 accept crypto, and RunPod asks for KYC before the first crypto payment. The [crypto comparison](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") lists them all. ### Are prices per GPU or per instance? Per GPU. An instance with four H100s costs four times $2.32/hour, billed per minute. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. ### What happens when my GPU-hours run out? When the GPU-hours of a GPU run out, its instances stop and their disks are kept; a new block of hours lets you start them again. Unused GPU-hours and balance never expire, so a paused project keeps its credit. ### Why is a spot or marketplace GPU sometimes cheaper than yours? Because the provider or the host can take it back, or sets its own price for its own machine. On 29 October 2026, Nebius listed preemptible H100s from $0.79 per GPU-hour. Our reference is the on-demand price, where the GPU stays yours while you pay. ### Can I get my money back? If an instance fails to start, or its GPU does not work and support cannot fix it, the hours it used go back to your balance. Unused GPU-hours and balance never expire. Top-ups are not paid out unless the law requires it; the part added only to reach the minimum top-up comes back on request within 30 days of payment, in USDT or USDC, as long as it is unspent. The [refund policy](https://gpuwithcrypto.com/refunds/ "Refund policy") has the details. ## Sources 1. [Massed Compute — VM pricing (H100, H200, B200, L40S references)](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Thunder Compute — Pricing (A100 80GB reference)](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 3. [SaladCloud price changes, November 2026 (RTX 4090 and RTX 5090 references)](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 4. [Runpod — Pricing (Secure Cloud and Community Cloud)](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 5. [Nebius — Prices (on-demand and preemptible)](https://nebius.com/prices) — Nebius (read 2026-10-29) 6. [Vast.ai public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 7. [Voltage Park — Pricing](https://www.voltagepark.com/pricing) — Voltage Park (read 2026-10-29) 8. [DigitalOcean — GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 9. [Jarvislabs — Pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 10. [Runpod — Billing information (crypto and KYC)](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Privacy policy > GPUwithCrypto privacy policy: no ID or card, short-lived billing lines, files on your disk not read, SSH private keys never seen, and no sale of data. - URL: https://gpuwithcrypto.com/privacy/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) A rented GPU holds your code, your data and your model weights. This page says exactly what we record, what we never collect, what we do not look at, and when anyone else gets to see a record. **Quick answer — What does GPUwithCrypto know about me?** Very little. **GPUwithCrypto** needs no ID, no selfie, no card and no email: you sign in with an account number. For billing it records each instance's GPU model, host, start and stop time, and it does not read the files on your instance's disk. BTC, USDT, USDC, ETH, SOL and LTC payments are public on their blockchains; a Monero payment shows neither sender nor amount there, but our billing still links it to your account. Records go to an authority only when a law compels us. ## 1. Who is responsible for your data? SerpGoat LLC, which operates GPUwithCrypto. Questions about this policy or about your records go to support@gpuwithcrypto.com. ## 2. What do we record? Only what running and billing GPU instances requires: - **Your account number**, which is how you sign in. Sign in with an account number; an email address is optional. - **Top-ups and orders:** the amount, the coin, the network, the transaction ID and the blocks of GPU-hours you bought. - **One billing line per instance:** the GPU model, the host, and the start and stop time. This is how per-minute billing is calculated. - **The SSH public keys you add**, so that your instances accept your connection. - **Your email address, only if you add one**, and the emails you send to support. How long each record is kept is in section 9. ## 3. What do we never collect? - No identity documents, selfies or company papers: a prepaid balance does not need them. - No card numbers: payment is crypto only. - No SSH private keys. You add your SSH public key in the dashboard; we never see or store your private key. - No copy of the files on your instance's disk. - No advertising profile, and no sale of data to anyone. ## 4. Do you look at the files on my instance? No. We do not read the files on your disk: your code, datasets, model weights and outputs are not opened, copied or scanned by us. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect, so you know in which country your files are stored while the instance exists. ## 5. Are my crypto payments private? Mostly not on the blockchain itself. The networks behind BTC, USDT, USDC, ETH, SOL and LTC are public ledgers: anyone can see a transfer's addresses, amount and time. Monero is the exception: an explorer shows neither the sender nor the amount of an XMR transfer. We never publish which account a transfer funded, but our billing links every payment, XMR included, to its account, orders, IP addresses and the TXID declared. Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited. That check uses the sending address, not your name or any document. ## 6. What is the optional email used for? To recover your account and to answer you. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good. Low-balance warnings appear in the dashboard and the API, so an email address is not needed to run instances. ## 7. Does the website use cookies or trackers? The public pages of this site carry no advertising and load no third-party analytics scripts. The dashboard uses a cookie to keep you signed in. Payment pages show the amount and the deposit address, with no outside tracking. ## 8. When do you share data? Data is never sold, rented or traded. Records go to an authority only when a law compels us, and then only the records that exist. We do not read your disk, so we hold no copy of its content to hand over. Child sexual abuse material found on the service is reported to the authorities, as the [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") says. ## 9. How long do you keep records? For billing we record each instance’s GPU model, host, start and stop time; these lines are kept for 30 days. We do not read the files on your disk. Account number, orders and top-ups stay while the account exists, then only as long as accounting law asks. Authorities receive records only when a law compels us. After the account is closed, its account number, orders and top-ups are kept only for the period that accounting law requires, then deleted. The disk of an instance is wiped when you delete the instance. ## 10. How can I see or erase my data? Write to support@gpuwithcrypto.com to ask what we hold about your account or to close it. Give an instance ID or a transaction ID so support can find the records. Nobody from our team will ever ask for your account number, your SSH private key or your API keys, by email or in a chat. Before closing an account, use your balance or check the [refund policy](https://gpuwithcrypto.com/refunds/ "Refund policy"): it says which part of a balance can be paid out. ## 11. Can this policy change? Yes. The date at the top of this page is the date of the version in force. A change that reduces your privacy applies only to records created after it is published here. --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Refund policy > When GPUwithCrypto gives GPU-hours or money back: instances that fail to start, faulty GPUs, and the part of a top-up added only to reach the minimum. - URL: https://gpuwithcrypto.com/refunds/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Two situations send the hours an instance used back to your balance, and one part of a top-up can be paid back in stablecoins. Everything else stays on your balance, which never expires. **Quick answer — Can I get a refund on GPUwithCrypto?** Yes, in three cases. If a **GPUwithCrypto** instance fails to start, or its GPU does not work and support cannot fix it, the hours it used go back to your balance. The part of a crypto top-up added only to reach the $1 minimum comes back in USDT or USDC on request, if unspent. Hours used by a working instance are not refunded; balance and unused hours never expire. ## 1. What is the full rule? If an instance fails to start, or its GPU does not work and support cannot fix it, the hours it used go back to your balance. Unused GPU-hours and balance never expire. Top-ups are not paid out unless the law requires it; the part added only to reach the minimum top-up comes back on request within 30 days of payment, in USDT or USDC, as long as it is unspent. A person reviews each request. Once it is approved, the hours appear on your balance and can pay for any GPU straight away. ## 2. Case 1: the instance never starts You start an H100 instance. It never reaches a running state, or it runs but no SSH connection ever succeeds. You write to support with the instance ID and the error you see. Support looks at it first. If it cannot get the instance working, every minute charged for that instance goes back to your balance, and you can start a new one when a GPU is free. ## 3. Case 2: the GPU does not work Your A100 instance starts, but `nvidia-smi` shows no GPU, or the GPU reports memory errors and your jobs crash on it. You write to support with the instance ID and the error output. Support tries to fix it. If it cannot, the hours the instance used go back to your balance. A GPU that works but runs your job slower than you hoped is not faulty. If a model does not fit or runs slowly, the [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") helps you choose a better GPU for the next run. ## 4. Case 3: the minimum top-up was more than you needed Every block, even the 10-hour RTX 4090 block at $2.80, costs more than the $1 minimum, so a top-up that matches your cart leaves no difference. A difference exists only when a top-up was raised to reach the minimum; it stays on your balance. If you leave that difference unspent, ask for it within the time limit in the rule above and it comes back to you in USDT or USDC. If you spend part of it on more GPU time, only the unspent part can come back. ## 5. What is not refunded? - **GPU time used by a working instance**, including time it ran idle before you stopped it. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. - **Hours used by an instance stopped for breaking the [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy").** - **The rest of a balance, and unused GPU-hours.** They are not paid out unless the law requires it, but they never expire: balance pays for any GPU, and hours stay available for their GPU model. - **Lost data or interrupted jobs**, such as files on a deleted instance or a run stopped because the balance reached zero. - **Waiting for a busy GPU.** Nothing is charged until an instance starts, so there is nothing to refund. ## 6. What about my rights under consumer law? They apply on top of this policy. If the law of the country where you live gives you a right to withdraw from a purchase or to get money back, that right is respected, and nothing on this page reduces it. ## 7. How do I ask for a refund? Email support@gpuwithcrypto.com and say which case applies. Include: - **For case 1 or 2:** the instance ID, the GPU model, the time in UTC and the error output, for example from `nvidia-smi` or your SSH client. - **For case 3:** the transaction ID of the top-up, and whether you want USDT or USDC and on which network. Support confirms the receiving address with you before anything is sent. Nobody from our team will ever ask for your account number, your SSH private key or your API keys, by email or in a chat. A person answers support emails within 12 hours, seven days a week, and helps you start, size and stop your GPU instances. ## 8. Why do refunds go to the balance rather than to my wallet? Crypto transfers have no chargeback, and each new transfer costs a network fee. Hours on the balance cost nothing to move and can pay for any GPU the same minute. Coins leave our address in one case only, the minimum top-up difference of case 3, and wherever a law makes a payout compulsory. --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Which GPU clouds accept crypto? > 14 of the 33 GPU clouds we checked on 29 Oct 2026 accept crypto and 4 ask for ID in some cases. Coins, processors, KYC and prices to rent a GPU with crypto. - URL: https://gpuwithcrypto.com/rent-gpu-with-crypto/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Coins, processors, ID rules, card needs and refund rules of 33 GPU clouds, read on their own pages. **Quick answer — Which GPU clouds accept crypto?** According to GPUwithCrypto's check of 29 October 2026, 14 of the 33 GPU clouds we checked accept crypto, including RunPod, Vast.ai, Massed Compute, Salad, Spheron, Vultr and Akash. Four of the 33 ask for ID in some cases, and RunPod asks for KYC before the first crypto payment. GPUwithCrypto rents an H100 for $2.32/hour in crypto with no ID check. - **33** GPU clouds checked on 29 October 2026 - **14** accept crypto on their official pages - **4** ask for ID in some cases (RunPod, Hetzner, OVHcloud, Aethir) - **1** ties KYC to paying in crypto (RunPod) ## Which GPU clouds accept crypto, and on what terms? Every row was read on the provider's own docs, billing pages or terms on 29 October 2026. "No ID step documented" means the pages we read describe no ID check, not a promise that the provider never asks. "—" means the pages did not say. Prices are the lowest self-serve on-demand price per GPU-hour we found, with the GPU in brackets; spot and marketplace prices are left out. | Provider | GPUs listed | Lowest on-demand $/GPU-hr | Crypto | Coins / processor | ID / KYC | Card for auto-pay | Crypto refundable | Minimum top-up | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** (ours) | H100, H200, B200, A100 80GB, L40S, RTX 4090, RTX 5090 | $0.28 (RTX 4090); H100 $2.32 | Yes, crypto only | BTC, USDT, USDC, ETH, SOL, LTC, XMR, sent on-chain to an address shown at checkout | None: no ID, no selfie, no documents | No card at all | Not paid out; balance never expires ([refunds](https://gpuwithcrypto.com/refunds/ "Refund policy")) | $1 | | RunPod | H100, H200, B200, A100, L40S, RTX 4090, RTX 5090 and more | $0.59 (RTX A6000); H100 $2.89 | Yes | Coins not named; Crypto.com named in its Terms | KYC before the first crypto payment | Yes, auto-pay is card-only | No: credits non-refundable | $10 to start | | Vast.ai | H100, H200, B200, A100, L40S, RTX 4090, RTX 5090 (marketplace) | Marketplace only | Yes | BitPay, Crypto.com (Coinbase Commerce legacy) | No ID step documented | Billing FAQ: card and email verified before adding credit | No: crypto-bought credits not refunded | — | | Massed Compute | H100, H200, B200, A100, L40S, RTX 6000 Ada, RTX A6000 | $0.55 (RTX A6000); H100 $2.73 | Yes | BTC by wallet invoice; processor not named | No ID step documented (email check only) | Yes: no auto-recharge for crypto | — | Not stated for crypto | | Salad | RTX 4090, RTX 5090 | $0.33 (RTX 4090; $0.44 from 1 Nov 2026) | Yes | USDC, USDT, RENDER on Solana via Coinbase x402 | No ID step documented | Yes: crypto is for manual top-ups only | No: all purchases final | $5 | | Spheron | H100, H200, B200, A100, L40S, RTX 5090, MI300X | $0.78 (RTX 6000 Ada); H100 $5.86 | Yes | USDC on Ethereum, Optimism, BNB Chain, Base, Arbitrum, Solana via CopperX | No ID step documented | — | No self-serve refunds; credits never expire | $50 in crypto ($15 by card) | | Vultr | L40S, A100 80GB on-demand | $1.67 (L40S) | Yes | BitPay: BTC, BCH, ETH, DOGE, PAX, BUSD, LTC, USDC, GUSD | No ID step documented; payment method on file | — | — | — | | Paperspace / DigitalOcean | H100, H200, L40S, MI300X, RTX 6000 Ada | $1.57 (L40S); H100 $4.41 | Yes, limited availability | Stablecoins from 400+ wallets; processor not named | No ID step documented; payment details used to verify identity | — | — | $25 one-time authorization, credited | | Shadeform | Resells 30+ clouds | Marketplace only | Yes (secondary source) | USDC via Stripe (Stripe case study) | No ID step documented | — | — | — | | Akash Network | H100, A100, RTX 4090, RTX 5090 (provider bids) | Marketplace bids only | Yes | AKT and ACT on-chain via Console Air; managed Console takes cards only | No ID step documented; no account on Console Air | — | — | 1 AKT (Console Air) | | io.net | No public price list | — | Yes | USDC via Spherepay, $IO; cards via Stripe | No ID step documented; withdrawals reviewed manually | — | Unused funds return as IO Credits | $10 | | Render Network | Mainly 3D rendering | — | Yes | RENDER on Solana, native | No ID step documented | — | No: credits cannot be withdrawn | 1 credit (EUR 1) | | Aethir | Application form, no self-serve sign-up | — | Yes | ATH from a bound wallet, native | KYC before claiming or withdrawing ATH | — | Unused ATH after KYC | — | | Nosana | — | — | Yes | NOS on Solana (SOL for fees); card credits via Stripe | No ID step documented | — | — | — | | Golem Network | GPU availability not documented | — | Yes | GLM, Polygon recommended, native | No account documented | — | — | — | | Lambda | B200, H100, A100, A10, RTX A6000 | $1.09 (RTX A6000); H100 $3.29 | No | Major credit cards only | No ID step documented; VPN purchases flagged | Card only | n/a | No prepaid model | | Thunder Compute | H100, A100, L40, RTX A6000 | $0.35 (RTX A6000); A100 $1.09 | No | Saved card or preloaded credit | No ID step documented | Yes: auto-pay uses a saved card | n/a | — | | Jarvislabs | H200, H100, A100, L4 | $0.44 (L4) | No | Card via Stripe; bank transfer above $500 | No ID step documented | — | n/a | $10 | | Nebius | H100, H200, B200, L40S | $1.55 (L40S) | No | Card via Stripe; bank transfer for companies | No ID step documented | — | n/a | $25 | | Verda (formerly DataCrunch) | B200, H200, H100, A100, L40S | $0.82 (RTX A6000) | No | Card, SEPA or invoice, coupons | No ID step documented | Card auto top-up | n/a | Invoices from 10,000 EUR/USD | | Crusoe | H100, H200, A100, L40S, MI300X | $1.50 (L40S) | No | Non-prepaid credit card | No ID step documented | Card only | n/a | Postpaid | | OVHcloud | H100, H200, A100, L40S, A10, L4 | $1.00 (A10) | No | Cards, PayPal, prepaid points | Sometimes: photo ID, card photo and selfie within 48 h | — | n/a | — | | Hetzner | RTX PRO 4000 / 6000 Blackwell only | — | No | "We do not accept cryptocurrencies" | Sometimes: government ID or card prepayment before an order | Postpaid | n/a | Postpaid | | Hot Aisle | AMD MI300X | $2.99 (MI300X) | No | Card via Stripe | No ID step documented | Card only | n/a | — | | Prime Intellect | Prices behind login | — | No | Standard methods; others via support | No ID step documented | — | n/a | — | | AWS | H100, H200, B200, A100, L40S, L4, A10G | $0.80 (L4, whole VM) | No | Cards, ACH, Advance Pay, local methods | Phone and payment checks, no ID document | Postpaid | n/a | Postpaid | | Google Cloud | H100, H200, B200, A100, L4 | $0.71 (L4, whole VM) | No | Cards, bank account; PayPal in 3 countries | No ID document in the payment docs | Postpaid | n/a | Postpaid | | Microsoft Azure | H100 (only GPU we priced) | $12.29 (H100, 8-GPU VM ÷ 8) | No | Credit card, debit card, wire transfer | Phone and card checks, no ID document | Postpaid | n/a | Postpaid | | Hyperbolic | H100, H200, B200 | $3.19 (H100) | Unknown: old stablecoin page now redirects | — | No ID step documented; optional phone check | Yes: card stored only for Auto Top-Up | — | $5 | | TensorDock | Public price page dated 2024 | — | Unknown: billing page behind login | — | No ID step documented | — | No refund policy in its terms | $5 | | CoreWeave | H100, H200, B200, A100, L40S, L40 | $1.25 (L40, 8-GPU node ÷ 8) | Unknown: sales-led | — | Not published | — | — | Contract via sales | | Voltage Park / Lightning AI | H100 ("Contact for pricing") | — | Unknown: docs moved | — | Unknown | — | — | — | | Fluidstack | No public price list | — | Unknown | — | Not published | — | — | — | | Genesis Cloud | Site unreachable on 29 Oct 2026 | — | Unknown | — | Unknown | — | — | — | *Prices checked on October 29, 2026.* USD per GPU-hour, taxes excluded. Multi-GPU prices are divided by the GPU count. AWS, Google Cloud and Azure prices include the whole VM. Hyperbolic, Genesis Cloud, Voltage Park, CoreWeave and Fluidstack are flagged for a re-check. ## What are the best GPU clouds to rent with crypto? **Methodology.** Published by GPUwithCrypto (SerpGoat LLC). Every provider is compared on the same criteria, taken from its own public website: crypto accepted on official pages, ID or KYC rules, lowest self-serve on-demand price per GPU-hour, card needed for auto-pay, refund of crypto top-ups, minimum top-up, the date of the check. *Prices checked on October 29, 2026.* Eight crypto-accepting providers with at least one public GPU price, ranked for paying in crypto without an ID check. ### 1. GPUwithCrypto — Our service Seven on-demand NVIDIA GPUs, paid only in crypto. Each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. We do not claim to be the cheapest for every GPU; spot and marketplace offers (Vast.ai) can be lower. - Best for: Renting a whole on-demand GPU without a card or an ID check - Price: H100 $2.32/hour · A100 80GB $0.92/hour · RTX 4090 $0.28/hour - KYC: None: no ID, no selfie, no documents - Crypto payment: BTC, USDT, USDC, ETH, SOL, LTC, XMR - Pros: Billed per minute; unused GPU-hours never expire - Pros: Whole GPUs on demand, never spot - Pros: Account number login, email optional - Cons: No spot tier - Cons: Crypto only: no card, no PayPal - Cons: A lost account number cannot be recovered without an email on file ### 2. Massed Compute — Lowest on-demand H100 we found Listed the lowest self-serve on-demand H100 80GB, H200 ($3.62), B200 and L40S prices we found. - Best for: Bitcoin holders who want list-price H100 or L40S VMs - Price: H100 80GB $2.73 · L40S $0.97 per GPU-hour - KYC: No ID step documented; email verification only - Crypto payment: BTC by wallet invoice - Pros: "No Bandwidth Charges" on its pricing page - Pros: No long-term commitment - Cons: No auto-recharge for crypto: instances end about 12 hours after the balance goes negative - Cons: Bitcoin is the only coin documented ### 3. RunPod — All seven GPUs we track, on-demand Lists an on-demand price for all seven GPUs we track. - Best for: Renters who accept a KYC check and want many GPU types - Price: H100 PCIe $2.89 on-demand · Community Cloud $1.99 - KYC: KYC before the first crypto payment - Crypto payment: Yes; coins not named; Crypto.com - Pros: Per-second billing - Pros: New users can start with $10 - Cons: Credits are non-refundable - Cons: Auto-pay is card-only ### 4. Vast.ai — Lowest marketplace offers A marketplace of hosts. Its billing FAQ says credit is added once a credit card and an email are verified. - Best for: Short jobs where the lowest hourly offer matters more than a fixed host - Price: H100 SXM from $1.73 (Vast.ai public price feed, 29 Oct 2026) - KYC: No ID document required in its docs - Crypto payment: BitPay, Crypto.com - Pros: Per-second billing, no minimum hours - Pros: 40+ data centers listed - Cons: Credits bought with crypto are not refunded - Cons: The feed does not split on-demand and interruptible offers ### 5. Salad — Cheapest on-demand RTX 4090 Distributed consumer GPUs. Its High Priority tier listed the lowest on-demand RTX 4090 and RTX 5090 prices we found; both rise on 1 November 2026, to $0.44 and $0.60. - Best for: Image generation and batch jobs on RTX cards, paid in stablecoins - Price: RTX 4090 $0.33 · RTX 5090 $0.50 (until 31 Oct 2026) - KYC: No ID step documented - Crypto payment: USDC, USDT, RENDER on Solana - Pros: $5 minimum top-up - Pros: Per-second billing - Cons: Crypto payments cannot be reversed or refunded - Cons: Credits expire 12 months after purchase ### 6. Spheron Lists H100, H200, B200, A100, L40S and RTX 5090 from partner data centers, paid in USDC through CopperX. - Best for: USDC holders on Base, Arbitrum or Solana - Price: A100 80GB $1.69 · RTX 5090 $0.84 per GPU-hour - KYC: No ID step documented - Crypto payment: USDC on Ethereum, Optimism, BNB Chain, Base, Arbitrum, Solana - Pros: Credits never expire - Pros: Failed deployments are not charged - Cons: $50 minimum for a crypto deposit - Cons: 20-minute minimum runtime - Cons: Its pricing page ($5.86) and H100 page ("from $2.65/hr") disagreed on 29 October 2026 ### 7. Vultr Takes nine coins through BitPay. - Best for: Paying in Bitcoin, Litecoin or Dogecoin for an L40S or A100 - Price: L40S $1.67 · A100 80GB $2.40 per GPU-hour - KYC: No ID step documented; a payment method on file "verifies your identity" - Crypto payment: BTC, BCH, ETH, DOGE, PAX, BUSD, LTC, USDC, GUSD via BitPay - Cons: H100 and B200 plans not marked on-demand in its API ### 8. Akash Network A decentralised marketplace priced by provider bids. Console Air needs only a Keplr wallet and AKT. - Best for: Wallet-only deployments with no account - Price: H100 SXM bids from $2.04 per GPU-hour - KYC: None documented; no account on Console Air - Crypto payment: AKT and ACT, on-chain - Cons: Stock changes with each provider - Cons: The managed Console needs a card for H100-class GPUs *Disclosure: this page is published by GPUwithCrypto. Competitor details come from their public websites at the date shown above and may have changed since.* ## Does a GPU cloud that accepts crypto skip KYC? No. Accepting crypto and asking for ID are separate rules. RunPod's billing docs say: "Complete any required KYC verification before your first crypto payment." Aethir takes deposits in ATH but asks for KYC before you claim or withdraw it. Several clouds treat the payment method itself as the identity check. Vultr says a valid payment method "verifies your identity", and DigitalOcean uses payment details the same way. Payment processors such as BitPay and Crypto.com can add checks of their own; the provider pages we read do not describe them. At GPUwithCrypto, no KYC means no ID, no selfie, no documents and no card. The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") still applies, and Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited. The [no-KYC GPU cloud page](https://gpuwithcrypto.com/no-kyc-gpu-cloud/ "Is there a GPU cloud with no KYC?") lists who asks for ID and when. ## What happens to crypto you do not spend? At most GPU clouds, a crypto top-up is gone once it is credited: the "Crypto refundable" column above gives each rule. Two exceptions: Aethir pays out unused ATH after KYC, and io.net returns unused funds as IO Credits, with withdrawals reviewed by hand in 3 to 7 business days. At GPUwithCrypto: Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. So size the first top-up to the job; a small first payment tests the whole flow. ## How do I rent a GPU with crypto on GPUwithCrypto? Get an account number, top up at least $1 in BTC, USDT, USDC, ETH, SOL, LTC, XMR, then start a GPU once the payment is credited. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. GPU-hours come in blocks of 10 to 720 hours and never expire. Every GPU price is on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). For an H100 only, see [where to rent an H100 with crypto](https://gpuwithcrypto.com/rent-h100-with-crypto/ "Where can I rent an H100 and pay with crypto?"). The [step-by-step guide](https://gpuwithcrypto.com/guides/rent-a-gpu-for-ai/ "How do I rent a GPU for AI, step by step?") covers SSH and the first job. ## Questions about renting GPUs with crypto ### Can I rent a GPU and pay with crypto? Yes. 14 of the 33 GPU clouds we checked on 29 October 2026 accept crypto. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no ID; RunPod, Vast.ai, Massed Compute, Salad and Spheron also take crypto, each with its own rules. ### Which GPU cloud accepts Bitcoin? Massed Compute documents Bitcoin through a wallet invoice, and Vultr takes it through BitPay. RunPod and Vast.ai do not name the coins they take. GPUwithCrypto accepts BTC, USDT, USDC, ETH, SOL, LTC, XMR. ### Can I pay for a cloud GPU with USDT or USDC? Yes. Salad takes USDC and USDT on Solana, Spheron takes USDC on six chains, and io.net takes USDC. The decentralised networks use their own tokens instead: AKT on Akash, RENDER on Render Network, NOS on Nosana. GPUwithCrypto accepts BTC, USDT, USDC, ETH, SOL, LTC, XMR. ### Does paying in crypto mean no KYC? Not always. RunPod asks for KYC before the first crypto payment, and Aethir before you withdraw unused ATH. GPUwithCrypto asks for no ID, no selfie and no documents. ### Can I get a refund on a crypto GPU top-up? Rarely. RunPod, Vast.ai, Salad and Render Network do not refund crypto credits, and Aethir pays back unused ATH only after KYC. At GPUwithCrypto, the balance never expires; the [refund policy](https://gpuwithcrypto.com/refunds/ "Refund policy") explains the one case where part of a top-up comes back. ## Sources 1. [Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 2. [Terms of service](https://www.runpod.io/legal/terms-of-service) — RunPod (read 2026-10-29) 3. [Billing reference](https://docs.vast.ai/documentation/reference/billing) — Vast.ai (read 2026-10-29) 4. [Billing FAQ](https://docs.vast.ai/guides/reference/faq/billing.md) — Vast.ai (read 2026-10-29) 5. [Crypto: manually add credits](https://vm-docs.massedcompute.com/docs/billing/crypto/manually-add-credits) — Massed Compute (read 2026-10-29) 6. [Crypto: automatic payments](https://vm-docs.massedcompute.com/docs/billing/crypto/automatic-payments) — Massed Compute (read 2026-10-29) 7. [Billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 8. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 9. [Billing](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 10. [What payment methods do you accept?](https://docs.vultr.com/support/platform/billing/what-payment-methods-do-you-accept) — Vultr (read 2026-10-29) 11. [Manage payment methods](https://docs.digitalocean.com/platform/billing/manage-payment-methods/) — DigitalOcean (read 2026-10-29) 12. [Shadeform customer story](https://stripe.com/customers/shadeform) — Stripe (read 2026-10-29) 13. [Console Air](https://akash.network/docs/developers/deployment/console-air) — Akash Network (read 2026-10-29) 14. [IO Credits](https://io.net/docs/guides/payment/io-credits.md) — io.net (read 2026-10-29) 15. [Render Credits](https://know.rendernetwork.com/docs/web-app/render-credits.md) — Render Network (read 2026-10-29) 16. [Manage wallet](https://docs.aethir.com/aethir-cloud/aethir-cloud-customer/cloud-customer-portal-guide/manage-wallet) — Aethir (read 2026-10-29) 17. [Wallet](https://learn.nosana.com/about/wallet) — Nosana (read 2026-10-29) 18. [Requestor FAQ](https://docs.golem.network/docs/golem/overview/requestor-faq) — Golem Network (read 2026-10-29) 19. [Manage billing](https://docs.lambda.ai/public-cloud/manage-billing/) — Lambda (read 2026-10-29) 20. [Billing](https://www.thundercompute.com/docs/billing.md) — Thunder Compute (read 2026-10-29) 21. [Settings and billing](https://docs.jarvislabs.ai/settings/) — Jarvislabs (read 2026-10-29) 22. [Sign up](https://docs.nebius.com/signup-billing/sign-up) — Nebius (read 2026-10-29) 23. [Pricing and billing](https://docs.verda.com/welcome-to-verda/pricing-and-billing/) — Verda (read 2026-10-29) 24. [Create an account](https://docs.crusoe.ai/create-an-account/index.html) — Crusoe (read 2026-10-29) 25. [Payment terms](https://www.ovhcloud.com/en/terms-and-conditions/payment/) — OVHcloud (read 2026-10-29) 26. [Fraud prevention FAQ](https://docs.hetzner.com/general/security-and-identify/fraud-prevention-faq/) — Hetzner (read 2026-10-29) 27. [Home page](https://hotaisle.xyz/) — Hot Aisle (read 2026-10-29) 28. [FAQ](https://docs.primeintellect.ai/faq.md) — Prime Intellect (read 2026-10-29) 29. [Getting started with an AWS account](https://docs.aws.amazon.com/accounts/latest/reference/getting-started.html) — AWS (read 2026-10-29) 30. [Payment methods](https://docs.cloud.google.com/billing/docs/how-to/payment-methods) — Google Cloud (read 2026-10-29) 31. [Troubleshoot Azure sign-up](https://learn.microsoft.com/en-us/azure/cost-management-billing/troubleshoot-subscription/troubleshoot-azure-sign-up) — Microsoft (read 2026-10-29) 32. [Billing and payments](https://hyperbolic.ai/docs/general/billing-payments.md) — Hyperbolic (read 2026-10-29) 33. [Terms of service](https://docs.tensordock.com/legal-information/terms-of-service-tos.md) — TensorDock (read 2026-10-29) 34. [Contact us](https://www.coreweave.com/contact-us) — CoreWeave (read 2026-10-29) 35. [Payment methods (legacy page)](https://docs.voltagepark.com/voltage-park-on-demand/billing/payment-methods) — Voltage Park (read 2026-10-29) 36. [Home page](https://www.fluidstack.io/) — Fluidstack (read 2026-10-29) 37. [Home page (unreachable)](https://www.genesiscloud.com/) — Genesis Cloud (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # How can I rent a GPU without a credit card? > How to rent a GPU without a credit card: pay in crypto, PayPal or bank transfer. Payment methods of 23 GPU clouds, checked 29 Oct 2026, and card-decline fixes. - URL: https://gpuwithcrypto.com/rent-gpu-without-credit-card/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Most GPU clouds want a credit card, and several refuse debit, prepaid or virtual cards. Crypto, PayPal and bank transfer are the three ways around it; here is who takes which. **Quick answer — How can I rent a GPU without a credit card?** To rent a GPU without a credit card, pay in crypto, or use a cloud that takes PayPal or bank transfer. **GPUwithCrypto** needs no card at any step: top up $1 or more in crypto and rent an H100 for $2.32/hour. As of 29 October 2026, 14 of the 33 GPU clouds we checked accept crypto and 5 take PayPal. ## Which GPU clouds take payment without a credit card? According to GPUwithCrypto's check of 29 October 2026, 14 of the 33 GPU clouds we checked accept crypto and 5 take PayPal, while Lambda, Crusoe and Hot Aisle mention cards only. The table lists the 23 providers whose payment pages name their methods. "—" means the pages we read did not say. | Provider | Card | PayPal | Crypto | Bank transfer / other | Card rules | | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** (ours) | No | No | Yes: BTC, USDT, USDC, ETH, SOL, LTC, XMR | No | No card at any step | | RunPod | Yes (Stripe) | — | Yes, after KYC | Invoicing above $5,000 via sales | Prepaid cards: $100 minimum per transaction | | Vast.ai | Yes | — | Yes: BitPay, Crypto.com | — | Billing FAQ: card and email verified before credit is added | | Massed Compute | Yes | — | Yes: BTC | — | Auto-recharge runs on a card only | | Salad | Credit or debit | — | Yes: USDC, USDT, RENDER on Solana | Above $10,000 per transaction, via sales | — | | Spheron | Yes (Stripe), from $15 | — | Yes: USDC, from $50 | — | — | | Paperspace / DigitalOcean | Yes | Yes ($5 authorization) | Stablecoins, limited availability | — | No virtual, electron or prepaid cards | | Vultr | — | — | Yes: BitPay | — | A payment method on file before deploying | | io.net | Yes (Stripe) | — | Yes: USDC, $IO | — | One docs page lists a 5% fee on card top-ups | | Akash Network | Managed Console only | — | Yes: AKT, ACT via Console Air | — | Managed Console needs a card for H100-class GPUs | | Render Network | Credits via Stripe | Credits via PayPal | Yes: RENDER | — | — | | Nosana | Credits via Stripe | — | Yes: NOS | — | — | | Lambda | Major credit cards only | No | No | No | Debit and prepaid cards declined | | Crusoe | Non-prepaid credit card | — | No | — | Card needed before any resource starts | | Hot Aisle | Yes (Stripe) | — | No | — | Card is the only method mentioned | | Thunder Compute | Saved card, or preloaded credit | — | No | — | — | | Jarvislabs | Yes (Stripe) | — | No | Above $500, via support | — | | Nebius | Yes (Stripe) | — | No | Companies, via sales | — | | Verda (formerly DataCrunch) | Yes, auto top-up | — | No | SEPA or invoice, from 10,000 EUR/USD | — | | Hetzner | Visa, Mastercard, AMEX, UnionPay | Yes | No | SEPA direct debit (EUR), bank transfer | Can ask for ID or a card prepayment | | OVHcloud | Visa, Mastercard, Maestro | Yes | No | Prepaid points account | Orders can be held for documents | | AWS | Yes | — | No | ACH direct debit, Advance Pay, local methods | Phone verification at sign-up | | Google Cloud | Credit, or Visa/Mastercard debit | Spain, Italy, Mexico only | No | Bank account where supported | No prepaid cards | | Microsoft Azure | Credit or debit | — | No | Wire transfer for approved businesses; SEPA in Europe | No virtual or prepaid cards | *Prices checked on October 29, 2026.* Read on each provider's own billing or payment pages on 29 October 2026. Hetzner's GPU line had no GPU from our list (RTX PRO 4000 and 6000 Blackwell only). ## Which GPU clouds require a credit card? Three mention a card as the only way to pay: Lambda ("only accepts major credit cards"), Crusoe (a non-prepaid credit card before any resource starts) and Hot Aisle (card through Stripe). Others need a card for part of the service: - **Vast.ai**: its billing FAQ adds credit once a card and an email are verified, even though it also takes crypto. Details: [does Vast.ai accept crypto?](https://gpuwithcrypto.com/does-vast-ai-accept-crypto/ "Does Vast.ai accept crypto?") - **Akash Network**: the managed Console needs a card for H100-class GPUs; Console Air needs only a wallet. - **Massed Compute** and **RunPod**: crypto works for manual top-ups, but automatic payments run on a card only. RunPod also asks for KYC first: [does RunPod accept crypto?](https://gpuwithcrypto.com/does-runpod-accept-crypto/ "Does RunPod accept crypto?") - **Microsoft Azure**: its sign-up calls the card check "identity verification". ## Why was my card declined by a GPU cloud? Usually because of one of the rules below, all taken from the providers' own pages on 29 October 2026. - **Card type.** Lambda declines debit and prepaid cards. DigitalOcean refuses virtual, electron and prepaid cards. Google Cloud and Azure refuse prepaid cards, and Crusoe wants a non-prepaid credit card. RunPod takes prepaid cards only from $100 per transaction. - **Region.** Lambda sells only in supported regions: the US, UK, Canada, Japan, Singapore and most of the EU, with Lithuania excluded. - **VPN.** Lambda flags purchases made over a VPN as suspicious. Hetzner advises against a VPN when you register. - **Phone.** Azure rejects VoIP numbers, and AWS needs a number that can receive a call or SMS. - **Fraud review.** Hetzner can ask for a government ID or a card prepayment before an order. OVHcloud can hold an order until you send documents within 48 hours. - **Verification hold.** Lambda places a $10 pre-authorization, refunded within days. Azure places a small temporary hold. If the card still fails, crypto avoids the card network altogether. The [GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?") are compared coin by coin. ## Can I rent a GPU with PayPal? Yes, at a few clouds. OVHcloud takes PayPal and listed an H100 at $2.99 per hour. DigitalOcean (Paperspace) takes PayPal after a $5 authorization charge. Google Cloud takes PayPal in Spain, Italy and Mexico only. Hetzner takes PayPal, but its GPU servers had no GPU from our list. Render Network sells credits through PayPal. Neither RunPod's nor Vast.ai's billing pages we read list PayPal. ## Can I pay for a cloud GPU by bank transfer? Yes, but mostly for larger or business accounts. Verda takes invoice top-ups from 10,000 EUR or USD. Jarvislabs takes bank transfers above $500 through support. Nebius takes bank transfers from companies through sales, and RunPod invoices business customers above $5,000. Hetzner offers SEPA direct debit and bank transfer, AWS offers ACH direct debit, and Azure offers wire transfer to approved businesses. For a single GPU for a few hours, crypto is the shortest path without a card. Our per-hour prices for each GPU are on [GPU pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## How do I rent a GPU with crypto instead of a card? On GPUwithCrypto the whole flow runs without a card, a bank account or an ID. Once your payment is confirmed, you launch an instance from the dashboard; our team prepares it at the GPU provider and its SSH details appear there as soon as it runs. Billing starts only then. 1. **Top up $1 or more** — Choose the amount and the coin (BTC, USDT, USDC, ETH, SOL, LTC, XMR). The crypto amount and its rate are locked for 30 minutes. 2. **Send it to the address shown, on the exact network** — The payment page shows our deposit address for the coin and network you picked: BTC on Bitcoin, USDT on Tron (TRC-20) or Ethereum (ERC-20), USDC on Ethereum (ERC-20), ETH on Ethereum, SOL on Solana, LTC on Litecoin, XMR on Monero. Send only that coin on that network: funds sent on another network or in another coin can’t be recovered. The network fee comes on top of the amount. 3. **Send the exact amount shown, then click “I’ve sent the payment”** — Each top-up has its own exact amount (down to the last decimals): that is how we find your payment and link it to your account. 4. **Credited once your payment is checked on the blockchain** — Your balance is credited once we have checked the transaction on the blockchain, after the network has confirmed it. Typical network confirmation times: 5. **Start your GPU** — Pick a block of GPU-hours first and the top-up is set to its price, or to the minimum for a cheaper block. Once the payment is confirmed, launch an instance from the dashboard: its SSH details appear as soon as it runs. Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. | Crypto | Network | Confirmations | Usual time | | --- | --- | --- | --- | | Bitcoin (BTC) | Bitcoin | 2 | ≈ 20 min | | Tether (USDT) | Tron (TRC-20) | 20 | ≈ 1 min | | Tether (USDT) | Ethereum (ERC-20) | 12 | ≈ 5 min | | USD Coin (USDC) | Ethereum (ERC-20) | 12 | ≈ 5 min | | Ethereum (ETH) | Ethereum | 12 | ≈ 5 min | | Solana (SOL) | Solana | 1 | ≈ 1 min | | Litecoin (LTC) | Litecoin | 6 | ≈ 15 min | | Monero (XMR) | Monero | 10 | ≈ 20 min | Closed the tab? The pending top-up stays under Billing in your account: open it to see the address and the exact amount again. ## Paying for a GPU without a card: questions ### Can I rent a GPU with a prepaid card? Rarely. Lambda, DigitalOcean, Google Cloud, Azure and Crusoe refuse prepaid cards. RunPod accepts them from $100 per transaction. ### Can I rent a GPU with a debit card? At some clouds. Lambda declines debit cards, while Google Cloud takes Visa and Mastercard debit, Azure takes debit cards and Salad takes credit or debit cards. ### Can I rent a GPU without a credit card in the UK or Canada? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR with no card from the UK, Canada and most other countries. Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine. ### Do I need a bank account to rent a GPU with crypto? Not on GPUwithCrypto. You need a wallet holding BTC, USDT, USDC, ETH, SOL, LTC, XMR, and at least $1 for the first top-up. ### Can I use a card on GPUwithCrypto? No. GPUwithCrypto is crypto only: no card, no PayPal, no bank transfer. That is why it asks for no card details and no ID. ## Sources 1. [Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 2. [Billing FAQ](https://docs.vast.ai/guides/reference/faq/billing.md) — Vast.ai (read 2026-10-29) 3. [Billing overview](https://vm-docs.massedcompute.com/docs/billing/overview) — Massed Compute (read 2026-10-29) 4. [Billing](https://docs.salad.com/general/explanation/billing) — Salad (read 2026-10-29) 5. [Billing](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 6. [Manage payment methods](https://docs.digitalocean.com/platform/billing/manage-payment-methods/) — DigitalOcean (read 2026-10-29) 7. [What payment methods do you accept?](https://docs.vultr.com/support/platform/billing/what-payment-methods-do-you-accept) — Vultr (read 2026-10-29) 8. [io.cloud payments](https://io.net/docs/guides/payment/io-cloud-payments.md) — io.net (read 2026-10-29) 9. [Quick start](https://akash.network/docs/getting-started/quick-start/) — Akash Network (read 2026-10-29) 10. [Render Credits](https://know.rendernetwork.com/docs/web-app/render-credits.md) — Render Network (read 2026-10-29) 11. [Payments API](https://learn.nosana.com/api/payments) — Nosana (read 2026-10-29) 12. [Manage billing](https://docs.lambda.ai/public-cloud/manage-billing/) — Lambda (read 2026-10-29) 13. [Create an account](https://docs.crusoe.ai/create-an-account/index.html) — Crusoe (read 2026-10-29) 14. [Home page](https://hotaisle.xyz/) — Hot Aisle (read 2026-10-29) 15. [Billing](https://www.thundercompute.com/docs/billing.md) — Thunder Compute (read 2026-10-29) 16. [Settings and billing](https://docs.jarvislabs.ai/settings/) — Jarvislabs (read 2026-10-29) 17. [Pay by card](https://docs.nebius.com/signup-billing/payments/card) — Nebius (read 2026-10-29) 18. [Pricing and billing](https://docs.verda.com/welcome-to-verda/pricing-and-billing/) — Verda (read 2026-10-29) 19. [Payment overview](https://docs.hetzner.com/general/billing-and-account-management/billing-at-hetzner/payment-overview) — Hetzner (read 2026-10-29) 20. [Payment terms](https://www.ovhcloud.com/en/terms-and-conditions/payment/) — OVHcloud (read 2026-10-29) 21. [Managing your payments](https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/manage-payments.html) — AWS (read 2026-10-29) 22. [Payment methods](https://docs.cloud.google.com/billing/docs/how-to/payment-methods) — Google Cloud (read 2026-10-29) 23. [Troubleshoot Azure sign-up](https://learn.microsoft.com/en-us/azure/cost-management-billing/troubleshoot-subscription/troubleshoot-azure-sign-up) — Microsoft (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Where can I rent an H100 and pay with crypto? > Where to rent an H100 and pay with crypto: 4 of the 16 on-demand H100 clouds we checked on 29 Oct 2026 accept crypto. Prices, coins and ID rules compared. - URL: https://gpuwithcrypto.com/rent-h100-with-crypto/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Most H100 clouds want a card. Four of the 16 with a public on-demand H100 price take crypto, one of them asks for KYC first, and three marketplaces take crypto for H100 time too. **Quick answer — Where can I rent an H100 and pay with crypto?** You can rent an H100 and pay with crypto at **GPUwithCrypto** for $2.32/hour, with no ID check. As of 29 October 2026, 4 of the 16 providers with a self-serve on-demand H100 price accept crypto: Massed Compute (Bitcoin, $2.73), RunPod ($2.89, KYC before the first crypto payment), Paperspace/DigitalOcean ($4.41) and Spheron ($5.86). ## Which providers rent an H100, and which take crypto? According to GPUwithCrypto's price check of 29 October 2026, 16 providers listed a self-serve on-demand H100 80GB price. Four of them accept crypto, and only RunPod ties a KYC check to the crypto payment. The price is the lowest on-demand H100 row per provider, in USD per GPU-hour. | Provider | H100 on-demand $/GPU-hr (variant) | Spot / marketplace | Crypto | ID / KYC | Card for auto-pay | | --- | --- | --- | --- | --- | --- | | **GPUwithCrypto** (ours) | **$2.32/hour** (H100 80GB, whole GPU) | — (on-demand only) | BTC, USDT, USDC, ETH, SOL, LTC, XMR | None | No card at all | | Massed Compute | $2.73 (80GB, form factor not stated); SXM $3.14 | — | Yes: BTC | No ID step documented | Yes: no auto-recharge for crypto | | RunPod | $2.89 (PCIe); SXM $3.99 | Community Cloud $1.99 (PCIe) | Yes: via Crypto.com | KYC before the first crypto payment | Yes: auto-pay is card-only | | OVHcloud | $2.99 (form factor not stated) | — | No | Sometimes: documents review | — | | Hyperbolic | $3.19 (SXM) | — | Unknown | No ID step documented | Card for Auto Top-Up | | Thunder Compute | $3.20 (SXM) | — | No | No ID step documented | Saved card | | Lambda | $3.29 (PCIe); SXM $3.99 in 8-GPU nodes | — | No | No ID step documented | Card only | | Jarvislabs | $3.49 (SXM) | Spot $1.54 | No | No ID step documented | — | | Crusoe | $3.90 (SXM) | — | No | No ID step documented | Card only | | Paperspace / DigitalOcean | $4.41 (SXM) | — | Yes: stablecoins, limited availability | No ID step documented | — | | Nebius | $4.50 (SXM) | Preemptible from $0.79 | No | No ID step documented | — | | Verda (formerly DataCrunch) | $4.75 (SXM) | Spot $2.37 | No | No ID step documented | Card auto top-up | | Spheron | $5.86 (form factor not stated) | $2.79 | Yes: USDC | No ID step documented | — | | CoreWeave | $6.16 (SXM, 8-GPU node ÷ 8) | Spot $2.46 | Unknown | Not published | — | | AWS | $6.88 (8-GPU VM ÷ 8, whole VM) | — | No | Phone and payment checks | Postpaid | | Google Cloud | $11.06 (SXM, 8-GPU VM ÷ 8, whole VM) | Spot $5.70 | No | No ID document | Postpaid | | Microsoft Azure | $12.29 (8-GPU VM ÷ 8, whole VM) | Spot $2.27 | No | Phone and card checks | Postpaid | | Vast.ai | Marketplace only | From $1.73 (SXM, Vast.ai public price feed) | Yes: BitPay, Crypto.com | No ID document required | Card and email verified before adding credit | | Akash Network | Marketplace only | Bids from $2.04 (SXM) | Yes: AKT, ACT | None documented | — | | Shadeform | Marketplace only | $2.50 (PCIe, lowest available listing) | Yes: USDC (secondary source) | No ID step documented | — | *Prices checked on October 29, 2026.* Taxes excluded. Multi-GPU prices are divided by the GPU count; AWS, Google Cloud and Azure include the whole VM. Not counted: Voltage Park's "starting at $1.99" (its pricing cards say "Contact for pricing"), Vultr's 8× H100 bare metal (not marked on-demand in its API) and TensorDock's 2024 price page. ## What is the cheapest H100 I can pay for in crypto? On-demand, GPUwithCrypto rents one for $2.32/hour, at least 15% under its named reference, Massed Compute at $2.73 on 29 October 2026. Marketplace offers can be lower: Vast.ai, which takes crypto, listed H100 offers from $1.73 in its public price feed. Among the other on-demand clouds that take crypto, the order was: 1. **Massed Compute**, $2.73, paid in Bitcoin. There is no auto-recharge for crypto, and instances end about 12 hours after the balance goes negative. 2. **RunPod**, $2.89 for a PCIe H100 and $3.99 for SXM, after KYC. 3. **Paperspace / DigitalOcean**, $4.41 for SXM, in stablecoins where its crypto option is available. 4. **Spheron**, $5.86 on its pricing page, in USDC. Its own H100 page said "from $2.65/hr" the same day. The median on-demand H100 across the 16 providers was $4.16 per GPU-hour, so the GPUwithCrypto price is 44% below the median. ## How much does an H100 cost on GPUwithCrypto? Buy a block of H100 hours, then pay per minute from it. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. | Plan | Quantity | Price | Price per unit | | --- | --- | --- | --- | | Try | 10 GPU-hours | $23.20 | $2.32/hour | | Day | 24 GPU-hours | $55.68 | $2.32/hour | | Week | 168 GPU-hours | $389.76 | $2.32/hour | | Month | 720 GPU-hours | $1,670.40 | $2.32/hour | In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire **Before you pay** - Minimum top-up $1. Top up once in crypto, then buy any plan from your balance. - Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. - Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min · XMR ≈ 20 min ## Is a marketplace or spot H100 cheaper? Yes, per hour, and several of them take crypto. On 29 October 2026, Vast.ai's public price feed listed an H100 SXM from $1.73 per hour, with 56 offers rentable. Akash provider bids started at $2.04, RunPod Community Cloud listed a PCIe H100 at $1.99, and Nebius quoted preemptible H100s from $0.79. The trade-off is control. A marketplace host sets its own terms, Vast.ai's feed does not split on-demand from interruptible offers, and a preemptible VM can be taken back. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. The [spot GPU explainer](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?") covers when an interruption is worth the saving. ## SXM, PCIe or NVL: which H100 am I renting? It depends on the provider, and the label matters for multi-GPU jobs. Prices differ by variant: RunPod listed a PCIe H100 at $2.89 and an SXM H100 at $3.99, and Massed Compute an NVL pair at $2.92 per GPU. Some clouds do not state the form factor at all. Massed Compute's $2.73 "H100 (80GB)" is one of them, and it is the reference for our price. On GPUwithCrypto, the variant is shown before you start: Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. For model sizes, see [how much VRAM you need](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?"). ## How do I rent an H100 with crypto here? 1. **Get an account number** — No ID, no selfie, no card. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good. 2. **Top up in crypto** — Send at least $1 in BTC, USDT, USDC, ETH, SOL, LTC, XMR. The balance is credited after the network confirmations. 3. **Start an H100** — Stock changes through the day: the dashboard shows which GPUs are free right now, and nothing is charged until an instance starts. Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. 4. **Stop it when the job ends** — Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. ## H100 and crypto: common questions ### How much does it cost to rent an H100 per hour? On 29 October 2026 the lowest self-serve on-demand H100 was $2.73 (Massed Compute) and the median was $4.16 across 16 providers. GPUwithCrypto charges $2.32/hour, or $55.68 for 24 GPU-hours. ### Can I rent an H100 with Bitcoin or USDT? Yes. Massed Compute takes Bitcoin for its H100 VMs, Spheron takes USDC, and GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR. RunPod and Vast.ai name their processors but not the coins. ### Do I need ID to rent an H100 with crypto? Not at GPUwithCrypto, Massed Compute or Spheron, whose pages describe no ID step. RunPod asks for KYC before the first crypto payment. The [no-KYC GPU cloud page](https://gpuwithcrypto.com/no-kyc-gpu-cloud/ "Is there a GPU cloud with no KYC?") lists every ID rule we found. ### Can I rent several H100s at once? Yes. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. Each GPU is billed at $2.32/hour while the instance runs. ### What can I not do on a rented H100? No cryptocurrency mining, no password or hash cracking, no attacks on other systems (DDoS, scanning, exploitation), no malware, no spam, no child sexual abuse material, no intimate or deceptive images of real people without their consent, no weapons development and no work for sanctioned people or companies. The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") applies to every account. ## Sources 1. [Pricing (VM)](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 2. [Crypto: manually add credits](https://vm-docs.massedcompute.com/docs/billing/crypto/manually-add-credits) — Massed Compute (read 2026-10-29) 3. [Pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 4. [Billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) 5. [GPU Droplets pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 6. [Manage payment methods](https://docs.digitalocean.com/platform/billing/manage-payment-methods/) — DigitalOcean (read 2026-10-29) 7. [Pricing](https://www.spheron.network/pricing/) — Spheron (read 2026-10-29) 8. [Billing](https://docs.spheron.ai/billing) — Spheron (read 2026-10-29) 9. [Public Cloud prices](https://www.ovhcloud.com/en/public-cloud/prices/) — OVHcloud (read 2026-10-29) 10. [GPU marketplace](https://www.hyperbolic.ai/marketplace) — Hyperbolic (read 2026-10-29) 11. [Pricing](https://www.thundercompute.com/pricing) — Thunder Compute (read 2026-10-29) 12. [Pricing](https://lambda.ai/pricing) — Lambda (read 2026-10-29) 13. [Pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 14. [Cloud pricing](https://www.crusoe.ai/cloud/pricing) — Crusoe (read 2026-10-29) 15. [Prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 16. [Pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 17. [Pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 18. [EC2 on-demand price list, US East (N. Virginia)](https://b0.p.awsstatic.com/pricing/2.0/meteredUnitMaps/ec2/USD/current/ec2-ondemand-without-sec-sel/US%20East%20(N.%20Virginia)/Linux/index.json) — AWS (read 2026-10-29) 19. [Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 20. [Azure Retail Prices API, East US](https://prices.azure.com/api/retail/prices?$filter=armRegionName%20eq%20'eastus'%20and%20serviceName%20eq%20'Virtual%20Machines') — Microsoft (read 2026-10-29) 21. [Public GPU pricing feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 22. [GPU prices API](https://console-api.akash.network/v1/gpu-prices) — Akash Network (read 2026-10-29) 23. [Instance types API](https://api.shadeform.ai/v1/instances/types) — Shadeform (read 2026-10-29) 24. [Pricing](https://www.voltagepark.com/pricing) — Voltage Park (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # Terms of service > GPUwithCrypto terms in plain English: prepaid crypto balance, per-minute billing, GPU-hour blocks, on-demand instances, disk wiping, refunds and liability. - URL: https://gpuwithcrypto.com/terms/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Eighteen short sections, each answering one question. They apply to the website, the dashboard, the API and every instance started from your account. **Quick answer — What do the GPUwithCrypto terms say in brief?** You top up a prepaid balance in crypto, at least $1 at a time, you buy blocks of GPU-hours from it, and **GPUwithCrypto** charges those hours per minute while your GPU instances run. Instances are on-demand and stop when the GPU-hours of their GPU run out; their disk is wiped when you delete them. The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") is part of these terms. The seller is SerpGoat LLC. No KYC is required. ## 1. Who agrees to these terms? The account holder, who may be a person or an organisation, and SerpGoat LLC, which operates GPUwithCrypto. Opening an account, sending a top-up or starting an instance means accepting these terms and the [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy"). You must be legally able to enter into a contract where you live. If you act for an organisation, you confirm that you may bind it. ## 2. What do we sell? Time on cloud GPU instances: NVIDIA H100 80GB, H200 141GB, B200, A100 80GB, L40S 48GB, GeForce RTX 4090 24GB and RTX 5090 32GB, rented by the minute. Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. Instances have 1 to 8 GPUs depending on the host. Before you start, the dashboard shows each host’s country, GPU count, form factor (SXM or PCIe) and interconnect. The GPU model, its hourly price and the disk size shown before you start are what the instance provides. ## 3. How does the account work? Sign in with an account number; an email address is optional. Your account number is how you sign in, so store it carefully. With an optional email on file you can recover it; with no email, a lost number is gone for good. Anything done with your account number, your API keys or the SSH keys you added is treated as done by you, so store them like passwords and replace a leaked key at once. Nobody from our team will ever ask for your account number, your SSH private key or your API keys, by email or in a chat. ## 4. How do I pay? Prices are in US dollars. You fund a prepaid balance in BTC, USDT, USDC, ETH, SOL, LTC, XMR, at least $1 each time. Send the exact amount on the network the payment page names; the balance is credited once the transfer has the required confirmations on the blockchain. Network fees are paid by the sender. Credit on your balance does not expire and buys blocks of GPU-hours for any GPU. When a top-up is only made to reach the minimum, the unspent part comes back on request within 30 days of payment, in USDT or USDC. The rest of a balance is not paid out unless the law requires it. ## 5. How is GPU time billed? Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. The [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?") lists the hourly rate of every GPU. GPU-hours are sold in blocks of 10, 24, 168 or 720 hours for one GPU model, at the same hourly rate; instances of that GPU draw from them per minute, and unused hours never expire. A block is bought from the balance at the price shown when you buy it, and only instances of that GPU model draw from it. ## 6. How do on-demand instances run? Once your payment is confirmed, you launch an instance from the dashboard; our team prepares it at the GPU provider and its SSH details appear there as soon as it runs. Billing starts only then. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. You start, stop and delete instances from the dashboard or the API. You alone decide what runs on them, within the [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy"). ## 7. What happens when my GPU-hours run out? When the GPU-hours of a GPU run out, its instances stop and their disks are kept; a new block of hours lets you start them again. A job that is still running at that moment is interrupted. Keep enough GPU-hours for the whole run, and save checkpoints regularly; a stop caused by used-up GPU-hours is not a fault of the service. ## 8. What happens to the data on an instance's disk? Every instance has local NVMe disk, sized for its GPU and shown before you start. The disk is wiped when you delete the instance, so copy your checkpoints out first. A deleted disk cannot be recovered by anyone, including us. We do not read the files on your disk, and we keep no backup of it. Copying your data in and out, and keeping your own copies, is your responsibility. ## 9. Is GPU availability guaranteed? No. Stock changes through the day: the dashboard shows which GPUs are free right now, and nothing is charged until an instance starts. A GPU model can be fully in use when you want it, and buying a block of hours does not reserve a machine. We do not promise an uptime figure. If an instance fails to start, or its GPU does not work and support cannot fix it, the refund rule in section 10 applies. ## 10. When do I get hours or money back? If an instance fails to start, or its GPU does not work and support cannot fix it, the hours it used go back to your balance. Unused GPU-hours and balance never expire. Top-ups are not paid out unless the law requires it; the part added only to reach the minimum top-up comes back on request within 30 days of payment, in USDT or USDC, as long as it is unspent. The [refund policy](https://gpuwithcrypto.com/refunds/ "Refund policy") applies these rules to worked cases. ## 11. What rights do consumers keep? If you are a consumer, the law of the country where you live may give you rights that a contract cannot remove, such as a right to withdraw from a purchase or a remedy for a faulty service. Those rights apply on top of these terms, and nothing in these terms reduces them. ## 12. Which rules of use apply? The [acceptable use policy](https://gpuwithcrypto.com/acceptable-use/ "Acceptable use policy") is part of these terms. In short: No cryptocurrency mining, no password or hash cracking, no attacks on other systems (DDoS, scanning, exploitation), no malware, no spam, no child sexual abuse material, no intimate or deceptive images of real people without their consent, no weapons development and no work for sanctioned people or companies. The licences of the models, datasets, images and software you run apply to you, and complying with them is your responsibility. You may build and sell your own products on your instances. Reselling raw access to our GPUs, as GPU rental, needs our written agreement. ## 13. Where is the service not available? Not available to people or companies in Cuba, Iran, North Korea, Syria, Russia, Belarus, China, Yemen or Western Sahara, or in the occupied regions of Ukraine. Incoming payments are checked against public sanctions lists of blockchain addresses wherever the network shows the sending address; a payment from a listed address is not credited. By using the service you confirm that you are not located in those places, that you are not on a sanctions or export-control restricted-party list, and that you do not act for anyone who is. Accounts or payments linked to those places or parties are refused or closed, and the law may require us to hold their funds. ## 14. When can you suspend or close an account? When an account breaks these terms or the acceptable use policy, when a payment turns out to be fraudulent or comes from a listed address, or when a law requires it. Reports go to abuse@gpuwithcrypto.com. An instance that breaks the acceptable use policy is stopped and the hours it used are not refunded; child sexual abuse material is reported to the authorities. We tell you the reason unless the law forbids it. You can close your account at any time by writing to support@gpuwithcrypto.com; the [privacy policy](https://gpuwithcrypto.com/privacy/ "Privacy policy") explains what is erased and what accounting law makes us keep. ## 15. What do you promise, and what not? We provide the service as it stands on the day you use it: the GPU model, the disk and the software image shown before you start. We do not promise that a given model, framework or workload fits on a GPU or runs at a given speed, that the service is never interrupted, or that a use you plan is legal where you are. That is for you to check. ## 16. How far does your liability go? As far as the law allows, our total liability for all claims is limited to the amount you paid us for the service the claim is about. We are not liable for indirect losses such as lost profit, lost data, a lost training run or a missed deadline. Nothing here limits a liability that the law does not allow us to limit. ## 17. Will prices or these terms change? Yes, and every change carries a date. Our prices follow the rule on the [pricing page](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?") and show the date of the price check they come from. A new price applies to GPU time used after it is published. GPU-hours already bought in a block stay hours: a later price change does not reduce them. The date at the top of this page is the date of the version in force. A change applies from the day it is published here; use before that date stays under the earlier version. ## 18. Which law applies, and how do I reach you? These terms are governed by the law of SerpGoat LLC's jurisdiction, except where the law of your country gives you protections that cannot be set aside. Questions about these terms go to support@gpuwithcrypto.com. Abuse reports go to abuse@gpuwithcrypto.com. --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What is a cloud GPU (GPU as a service)? > What is a cloud GPU? A data-center GPU rented by the hour over SSH or API. An on-demand H100 cost $2.73 to $12.29 an hour on 29 Oct 2026. Terms, prices, crypto. - URL: https://gpuwithcrypto.com/what-is/cloud-gpu/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) A cloud GPU lets you use an NVIDIA H100 or RTX 4090 for an hour without buying one. Here are the terms you will meet, how the pricing works, and what each GPU cost per hour on 29 October 2026. **Quick answer — What is a cloud GPU (GPU as a service)?** A cloud GPU, or GPU as a service, is a data-center graphics card you rent by the second, minute or hour and use remotely over SSH, a notebook or an API, instead of buying it. On 29 October 2026 an on-demand NVIDIA H100 cost from $2.73 to $12.29 per GPU-hour across the 16 providers we checked. **GPUwithCrypto** rents cloud GPUs paid in crypto, with no KYC. ## What do the cloud GPU terms mean? **Cloud GPU (GPU as a service)** — A GPU in a provider's data center that you rent for a period and reach over the network, usually with SSH, Jupyter or an API. **On-demand** — A fixed list price per hour. You start and stop the GPU when you want, and nobody can reclaim it while it runs. **Spot / preemptible** — Spare capacity sold at a discount that the provider can take back at short notice. See [what is a spot GPU](https://gpuwithcrypto.com/what-is/spot-gpu/ "What is a spot GPU, and is it worth it?"). **Marketplace GPU** — A GPU listed by an independent host on a platform such as Vast.ai or Akash, at the host's own price. **Serverless GPU** — You send requests or jobs instead of renting a machine; the platform starts GPU workers when needed and bills only while they run. **GPU-hour** — One GPU for one hour. An 8-GPU instance running for one hour uses 8 GPU-hours. **VRAM** — The memory on the GPU itself. It decides which model fits: 24 GB on an RTX 4090, 80 GB on an H100, 141 GB on an H200. **HBM** — High Bandwidth Memory, stacked next to the GPU chip. The H100 SXM's HBM3 moves 3,350 GB/s; the L40S's GDDR6 moves 864 GB/s. **SXM** — NVIDIA's socketed module for multi-GPU boards (HGX). An H100 SXM draws up to 700 W and links to its neighbours over NVLink at 900 GB/s. **PCIe** — A standard plug-in card. The H100 PCIe draws 350 W, moves 2,000 GB/s of memory bandwidth and can link to another card through an NVLink bridge at 600 GB/s. **NVLink** — NVIDIA's direct GPU-to-GPU link, much faster than PCIe: 900 GB/s per H100 or H200 SXM, 1,800 GB/s per B200. The L40S and RTX 4090 have none. **MIG** — Multi-Instance GPU: one data-center GPU split into isolated slices, each with part of the memory. Cheaper, but not a whole GPU. ## How much does a cloud GPU cost per hour? Self-serve on-demand prices per GPU-hour read from each provider's own pricing page on 29 October 2026. Multi-GPU instance prices are divided by their GPU count; spot and marketplace prices are left out. | GPU | VRAM | Lowest | Median | Highest | Providers | GPUwithCrypto | | --- | --- | --- | --- | --- | --- | --- | | [H100 80GB](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?") | 80 GB | $2.73 (Massed Compute) | $4.16 | $12.29 (Microsoft Azure) | 16 | **$2.32/hour** | | [H200 141GB](https://gpuwithcrypto.com/gpu-prices/h200/ "How much does an H200 cost per hour?") | 141 GB | $3.62 (Massed Compute) | $5.29 | $10.60 (Google Cloud) | 13 | **$3.07/hour** | | B200 | 180 GB | $5.43 (Massed Compute) | $8.50 | $14.24 (AWS) | 9 | **$4.61/hour** | | [A100 80GB](https://gpuwithcrypto.com/gpu-prices/a100/ "How much does an A100 cost per hour?") | 80 GB | $1.09 (Thunder Compute) | $2.23 | $5.07 (Google Cloud) | 13 | **$0.92/hour** | | [L40S 48GB](https://gpuwithcrypto.com/gpu-prices/l40s/ "How much does an L40S cost per hour?") | 48 GB | $0.97 (Massed Compute) | $1.67 | $3.15 (Spheron) | 11 | **$0.82/hour** | | [RTX 5090](https://gpuwithcrypto.com/gpu-prices/rtx-5090/ "How much does it cost to rent an RTX 5090 in the cloud?") | 32 GB | $0.50 (Salad) | $0.84 | $1.19 (RunPod) | 3 | **$0.42/hour** | | [RTX 4090](https://gpuwithcrypto.com/gpu-prices/rtx-4090/ "How much does it cost to rent an RTX 4090 per hour?") | 24 GB | $0.33 (Salad) | $0.61 | $0.89 (RunPod) | 2 | **$0.28/hour** | *Prices checked on October 29, 2026.* USD, taxes excluded. AWS, Google Cloud and Azure prices cover a whole VM (vCPUs, RAM, local SSD) divided by its GPU count. Salad's RTX 4090 price rises from $0.33 to $0.44 on 1 November 2026. ## How does cloud GPU pricing work? You pay per GPU per unit of time, plus storage, at a rate set by the price type. The details that change the bill: - **Billing increment.** On 29 October 2026, RunPod billed per second, Lambda, Thunder Compute and Jarvislabs per minute, Google Cloud and OVHcloud per second after a 1-minute minimum, DigitalOcean per second with a 5-minute minimum. - **Price type.** On-demand, spot, marketplace and reserved prices are different products. CoreWeave advertises up to 60% off on-demand for committed usage; Verda's reserved discounts run from 2% for one month to 25% for two years. - **What is bundled.** Hyperscalers sell whole VMs; smaller GPU clouds price the GPU and add CPU, RAM and disk. - **Storage.** RunPod, for example, charged $0.10 per GB per month for container disk. - **Prepaid or postpaid.** AWS, Google Cloud and Azure bill afterwards against a card or contract. RunPod, Salad and GPUwithCrypto sell prepaid credit. ## Why do cloud GPU prices vary so much? Because the same GPU is sold in different packages by very different companies. According to GPUwithCrypto's price check of 29 October 2026, the most expensive on-demand H100 we found (Microsoft Azure, $12.29 per GPU-hour as part of an 8-GPU VM) cost 4.5 times the cheapest (Massed Compute, $2.73). The median across 16 providers was $4.16. Hyperscalers price a whole VM with its CPUs, RAM and local SSD; specialised GPU clouds price the GPU and add the rest. Form factor matters too: an H100 SXM has 3,350 GB/s of memory bandwidth and NVLink; an H100 PCIe has 2,000 GB/s. ## How do I pay for a cloud GPU? Most clouds want a card; a minority take crypto. Of the 33 GPU providers we checked on 29 October 2026, 14 accept crypto and 4 ask for ID in some cases; RunPod asks for KYC before the first crypto payment. GPUwithCrypto takes only crypto: BTC, USDT, USDC, ETH, SOL, LTC, XMR, from $1, with no ID, no selfie and no card. Comparison: [GPU clouds that accept crypto](https://gpuwithcrypto.com/rent-gpu-with-crypto/ "Which GPU clouds accept crypto?"). ## Which cloud GPU should I choose? The cheapest one whose VRAM holds your model plus about 20%. A 7–8B model fits a 24 GB RTX 4090; a 70B model in FP8 needs a 141 GB H200; image models like FLUX.1 run on an RTX 4090 or 5090. The [VRAM guide](https://gpuwithcrypto.com/how-much-vram-do-i-need/ "How much VRAM do I need to run or fine-tune an LLM?") has the arithmetic, and [best GPU for LLM inference](https://gpuwithcrypto.com/best-gpu-for/llm-inference/ "What is the best GPU for LLM inference?") the picks by model size. Then follow [how to rent a GPU for AI](https://gpuwithcrypto.com/guides/rent-a-gpu-for-ai/ "How do I rent a GPU for AI, step by step?"). ## Cloud GPU questions ### What does GPU as a service mean? Renting GPU computing time from a provider instead of owning the hardware. You pay per second, minute or hour, reach the GPU over the network, and the provider handles power, cooling and replacement. ### How much does a cloud GPU cost per hour? On 29 October 2026, the lowest self-serve on-demand prices we found were $0.33 for an RTX 4090, $1.09 for an A100 80GB and $2.73 for an H100. Hyperscalers charged up to $12.29 per H100-hour. ### What is the difference between SXM and PCIe GPUs? SXM modules sit on multi-GPU boards with NVLink between them and run at higher power; PCIe cards plug into a standard slot. For the H100: 3,350 against 2,000 GB/s of memory bandwidth, and 700 W against 350 W. ### What is a serverless GPU? A service where you send requests or jobs and the platform runs them on GPUs it starts and stops for you, billing only for the time workers run. You do not manage a machine, but you also do not keep one between requests. ### Is a cloud GPU shared with other users? It depends on the provider: some sell MIG slices or time-shared GPUs. At GPUwithCrypto: Each instance gives you whole GPUs, never a time-shared slice or a MIG partition. You get root SSH access to Ubuntu with the NVIDIA driver, CUDA and PyTorch installed, or start from your own Docker image. ### Can I rent a cloud GPU with crypto and without ID? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR and asks for no ID and no card. Sign in with an account number; an email address is optional. ## Sources 1. [NVIDIA H100 specifications](https://www.nvidia.com/en-us/data-center/h100/) — NVIDIA (read 2026-10-29) 2. [NVIDIA H100 PCIe product brief](https://www.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf) — NVIDIA (read 2026-10-29) 3. [NVIDIA H200 specifications](https://www.nvidia.com/en-us/data-center/h200/) — NVIDIA (read 2026-10-29) 4. [NVIDIA HGX platform (B200)](https://www.nvidia.com/en-us/data-center/hgx/) — NVIDIA (read 2026-10-29) 5. [NVIDIA L40S specifications](https://www.nvidia.com/en-us/data-center/l40s/) — NVIDIA (read 2026-10-29) 6. [Azure Retail Prices API (ND H100 v5, East US)](https://prices.azure.com/api/retail/prices?$filter=armRegionName%20eq%20'eastus'%20and%20serviceName%20eq%20'Virtual%20Machines') — Microsoft (read 2026-10-29) 7. [Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 8. [Massed Compute pricing](https://vm.massedcompute.com/pricing) — Massed Compute (read 2026-10-29) 9. [RunPod pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 10. [CoreWeave pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 11. [Verda pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 12. [DigitalOcean GPU Droplet pricing](https://www.digitalocean.com/pricing/gpu-droplets) — DigitalOcean (read 2026-10-29) 13. [RunPod billing information](https://docs.runpod.io/references/billing-information) — RunPod (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. --- # What is a spot GPU, and is it worth it? > What is a spot GPU? Spare capacity sold 40% to 82% below on-demand that the provider can take back. H100 spot prices of 29 Oct 2026, checkpoint tips, crypto. - URL: https://gpuwithcrypto.com/what-is/spot-gpu/ - Last updated: 2026-10-29 - Publisher: GPUwithCrypto (https://gpuwithcrypto.com) Spot GPUs are the cheapest way to rent a data-center GPU, as long as your job survives losing it. Here is what the terms mean, what spot cost on 29 October 2026, and how to make a job restartable. **Quick answer — What is a spot GPU, and is it worth it?** A spot GPU is spare cloud GPU capacity sold below the on-demand price that the provider can take back at short notice. On 29 October 2026, H100 SXM spot or preemptible prices were 40% to 82% below on-demand at the five providers that list both. It is worth it for jobs that checkpoint and restart. **GPUwithCrypto** rents on-demand GPUs only, paid in crypto, so the GPU stays yours while your balance covers it. ## What do spot, on-demand and marketplace mean? **Spot GPU** — Spare data-center capacity sold at a discount. The provider can stop or reclaim it when it needs the GPU back. Also called preemptible or interruptible. **On-demand GPU** — A GPU at a fixed list price per hour, started when you want and kept until you stop it. Nobody else can reclaim it. The basics: [what is a cloud GPU](https://gpuwithcrypto.com/what-is/cloud-gpu/ "What is a cloud GPU (GPU as a service)?"). **Preemptible VM** — The name Nebius and Google Cloud use for spot-style capacity: cheaper, and stopped when the provider needs it. **Marketplace or community GPU** — A GPU listed by a third-party host on a platform such as Vast.ai, at the host's own price. Reliability depends on that host. **Reserved or committed GPU** — A discount for promising to pay for a period, from one month to several years. **Checkpoint** — A saved copy of a job's state (model weights, optimizer state, step count) that lets it resume after an interruption. ## How do spot, on-demand and marketplace GPUs compare? | | On-demand | Spot / preemptible | Marketplace / community | | --- | --- | --- | --- | | Price | Fixed list price | Discounted, often variable | Set by each host | | Can be taken back | No | Yes | Depends on the host and offer | | Who sets the price | The provider | The provider, by spare capacity | The host | | Best for | Inference servers, interactive work, deadlines | Batch jobs that checkpoint | Cheap experiments, flexible batch work | | Examples (29 October 2026) | GPUwithCrypto, Lambda, Massed Compute | Nebius, Google Cloud, Verda, Jarvislabs | Vast.ai, RunPod Community Cloud | ## How much cheaper is a spot H100? H100 SXM per GPU-hour at the five providers that list both an on-demand and a spot or preemptible price, read from their own pricing pages on 29 October 2026. Multi-GPU instance prices are divided by their GPU count. | Provider | On-demand | Spot / preemptible | Discount | Note | | --- | --- | --- | --- | --- | | Nebius | $4.50 | from $0.79 | 82% | Preemptible "from" price, changes with demand | | CoreWeave | $6.155 | $2.4638 | 60% | 8-GPU instance ÷ 8 | | Jarvislabs | $3.49 | $1.54 | 56% | Interruptible "Spot / hr" | | Verda (formerly DataCrunch) | $4.746 | $2.373 | 50% | "Spot price" | | Google Cloud | $11.0613 | $6.5876 | 40% | a3-highgpu-8g ÷ 8, whole VM included | | **GPUwithCrypto** | $2.32/hour (H100 80GB) | — | — | On-demand only | *Prices checked on October 29, 2026.* Marketplace prices are another product: the cheapest H100 offer on Vast.ai's marketplace was $1.73 per hour (Vast.ai public price feed), and RunPod's Community Cloud H100 SXM cost $2.69 against $3.99 on its Secure Cloud. The two platforms are compared in [RunPod vs Vast.ai](https://gpuwithcrypto.com/compare/runpod-vs-vast-ai/ "RunPod vs Vast.ai: which is cheaper and more reliable?"). ## Is spot cheaper than on-demand? Yes, by about half. According to GPUwithCrypto's price check of 29 October 2026, the median spot discount on an H100 SXM across those five providers was 56%. Spot at Nebius and Jarvislabs, and the cheapest Vast.ai marketplace offer, were all below our on-demand H100 price. Our price rule compares only with on-demand prices: Each GPU is priced at least 15% under a named reference: the on-demand price of the same GPU at a self-serve provider, read on 29 October 2026 (H100: Massed Compute at $2.73 per GPU-hour; each GPU's reference is named on its price page). Spot, interruptible, marketplace and contact-sales prices are not used as references, and some of them are lower. We do not claim to be the cheapest for every GPU. Every on-demand H100 price we read is on [H100 rental prices](https://gpuwithcrypto.com/gpu-prices/h100/ "How much does it cost to rent an H100 per hour?"). ## How is spot pricing set? By spare capacity, so it moves. Nebius says the highest rate of a preemptible VM is always one cent below the current on-demand rate; its table shows only the "from" price. Google Cloud labels its spot column "Current Spot pricing". Salad sells priority tiers instead: on 29 October 2026 its lowest (batch) priority RTX 4090 cost $0.16 per hour against $0.33 for high priority, which is never preempted. ## When is a spot GPU worth it? When losing the GPU costs you minutes, not a deadline. At a 56% discount you pay 0.44 of the on-demand rate, so spot stays cheaper until the work you redo after interruptions exceeds 127% of the useful work (1 ÷ 0.44 = 2.27). The real risk is time, not money. **Good fits:** batch inference over a dataset, hyperparameter sweeps, training that checkpoints every few minutes, rendering queues. **Poor fits:** an inference API that must stay up, interactive notebooks and ComfyUI sessions, long jobs without checkpoints, and multi-GPU jobs, where losing one GPU stops all of them. ## How do I make a training job survive a spot interruption? Save checkpoints often, keep them off the instance, and resume from the last one. With Hugging Face's Trainer, set a save interval and a cap on how many checkpoints to keep, then call `train(resume_from_checkpoint=True)`: it loads the last checkpoint in the output folder, with optimizer and scheduler states. Copy the output folder to storage that outlives the instance, and make each step safe to run twice. *resume_after_interruption.py* ```python from transformers import Trainer, TrainingArguments args = TrainingArguments( output_dir="run", save_strategy="steps", save_steps=200, # a checkpoint every 200 steps save_total_limit=3, # keep the 3 most recent ) trainer = Trainer(model=model, args=args, train_dataset=train_ds) # after a restart: continue from the last checkpoint in ./run trainer.train(resume_from_checkpoint=True) ``` ## Why are GPUwithCrypto instances on-demand only? So a long job or a live endpoint is never cut by someone else. Instances are on-demand, not spot: nobody can take your GPU back while your balance covers it, and the instance runs until you stop it. When the GPU-hours of a GPU run out, its instances stop and their disks are kept; a new block of hours lets you start them again. Each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026. For the H100, that reference is Massed Compute at $2.73, and our price is 44% below the median of 16 providers. All prices: [pricing](https://gpuwithcrypto.com/pricing/ "How much does a cloud GPU cost per hour at GPUwithCrypto?"). ## Spot GPU questions ### What does spot GPU mean? A spot GPU is spare capacity a cloud sells at a discount on condition that it can take the GPU back when it needs it. Google Cloud and Nebius call the same idea preemptible; Jarvislabs calls it interruptible. ### How much cheaper is a spot GPU? On 29 October 2026, H100 SXM spot or preemptible prices were 40% (Google Cloud) to 82% (Nebius, from-price) below on-demand at the five providers that list both, with a median of 56%. ### Can a spot GPU be taken back in the middle of training? Yes, that is the condition of the discount. Save checkpoints every few minutes to storage outside the instance and resume from the last one, so an interruption costs only the steps since that checkpoint. ### Should I use a spot GPU for ComfyUI or an inference API? No. An interruption ends the session or takes the endpoint down. Use an on-demand GPU for interactive and live work, and spot for batch jobs that restart cleanly. ### Can I pay for an on-demand GPU with crypto? Yes. GPUwithCrypto takes BTC, USDT, USDC, ETH, SOL, LTC, XMR from a $1 top-up, with no ID and no card. Instances are billed per minute at the hourly rate while they run, from the GPU-hours you bought for that GPU. Stopping an instance stops the charge at once. ## Sources 1. [Nebius prices](https://nebius.com/prices) — Nebius (read 2026-10-29) 2. [CoreWeave pricing](https://www.coreweave.com/pricing) — CoreWeave (read 2026-10-29) 3. [Jarvislabs pricing](https://jarvislabs.ai/pricing) — Jarvislabs (read 2026-10-29) 4. [Verda pricing](https://verda.com/pricing) — Verda (read 2026-10-29) 5. [Accelerator-optimized VM pricing](https://cloud.google.com/products/compute/pricing/accelerator-optimized) — Google Cloud (read 2026-10-29) 6. [RunPod pricing](https://www.runpod.io/pricing) — RunPod (read 2026-10-29) 7. [Vast.ai public GPU price feed](https://storage.googleapis.com/vast-public-gpu-pricing/gpu-pricing-public.json) — Vast.ai (read 2026-10-29) 8. [SaladCloud price changes, November 2026](https://blog.salad.com/saladcloud-price-changes-november-2026) — Salad (read 2026-10-29) 9. [Trainer and TrainingArguments](https://huggingface.co/docs/transformers/main_classes/trainer) — Hugging Face (read 2026-10-29) --- GPUwithCrypto: GPUwithCrypto rents cloud GPUs (NVIDIA H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090) by the minute, paid in crypto with no KYC; each GPU is priced at least 15% under a named on-demand reference read on 29 October 2026.