H100 80GB
Training and fine-tuning up to 70B parameters, fast LLM inference. GPUwithCrypto: $2.32/hour.
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 | $2.73 (Massed Compute) | $4.16 | 16 | $2.32/hour | $55.68 | $1,670.40 | $2,991.60 |
| H200 141GB | $3.62 (Massed Compute) | $5.29 | 13 | $3.07/hour | $73.68 | $2,210.40 | $3,808.80 |
| B200 180GB | $5.43 (Massed Compute) | $8.50 | 9 | $4.61/hour | $110.64 | $3,319.20 | $6,120.00 |
| A100 80GB | $1.09 (Thunder Compute) | $2.23 | 13 | $0.92/hour | $22.08 | $662.40 | $1,602.72 |
| L40S 48GB | $0.97 (Massed Compute) | $1.67 | 11 | $0.82/hour | $19.68 | $590.40 | $1,203.12 |
| RTX 4090 24GB | $0.33 (Salad) | $0.61 | 2 | $0.28/hour | $6.72 | $201.60 | $439.20 |
| RTX 5090 32GB | $0.50 (Salad) | $0.84 | 3 | $0.42/hour | $10.08 | $302.40 | $604.80 |
Prices checked on
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?
H200 141GB
Large-model inference and long contexts: a 70B model in FP16 fits on two GPUs. GPUwithCrypto: $3.07/hour.
B200 180GB
Frontier-scale training and FP4/FP8 inference. GPUwithCrypto: $4.61/hour.
A100 80GB
Fine-tuning and inference on a budget, the most rented data-center GPU. GPUwithCrypto: $0.92/hour.
L40S 48GB
Image and video generation, inference of models up to about 30B parameters. GPUwithCrypto: $0.82/hour.
RTX 4090 24GB
ComfyUI, Stable Diffusion, Flux and LoRA training at the lowest hourly price. GPUwithCrypto: $0.28/hour.
RTX 5090 32GB
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 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
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 has the full table, and the cheapest GPU cloud ranking 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.
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 and asks for no ID, no selfie and no card. The crypto GPU cloud comparison lists coins, processors and refund rules, and the no-KYC GPU cloud guide 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 (14)
- Massed Compute — VM pricing vm.massedcompute.com/pricing — Massed Compute (read 2026-10-29)
- Thunder Compute — Pricing thundercompute.com/pricing — Thunder Compute (read 2026-10-29)
- SaladCloud price changes, November 2026 blog.salad.com/saladcloud-price-changes-november-2026 — Salad (read 2026-10-29)
- Runpod — Pricing runpod.io/pricing — RunPod (read 2026-10-29)
- Lambda — Pricing lambda.ai/pricing — Lambda (read 2026-10-29)
- Nebius — Prices nebius.com/prices — Nebius (read 2026-10-29)
- Google Cloud — Accelerator-optimized VM pricing — Google Cloud (read 2026-10-29)
- AWS — EC2 on-demand price list, US East (N. Virginia) — Amazon Web Services (read 2026-10-29)
- DigitalOcean — GPU Droplets pricing digitalocean.com/pricing/gpu-droplets — DigitalOcean (read 2026-10-29)
- Vast.ai public GPU pricing feed — Vast.ai (read 2026-10-29)
- Akash Network — GPU prices API console-api.akash.network/v1/gpu-prices — Akash Network (read 2026-10-29)
- Runpod — Billing information docs.runpod.io/references/billing-information — RunPod (read 2026-10-29)
- Transformer Math 101 — EleutherAI (read 2026-10-29)
- NVIDIA DGX B200 — NVIDIA (read 2026-10-29)