Try
10 GPU-hours
$4.20
$0.42/hour
- Test your setup end to end
- Same GPU and image as a long run
- Unused hours never expire
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
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.
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
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?
| 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 has the formulas.
| Model and job | Memory needed | One RTX 5090 32GB? |
|---|---|---|
| Llama 3.1 8B inference, BF16 | ≈ 19.2 GB | ✓ |
| Llama 3.1 8B, BF16 weights plus a full 128K-token context with FP8 cache | ≈ 24.6 GB | ✓ |
| Llama 3.1 8B LoRA fine-tune | ≈ 19.2 GB | ✓ |
| FLUX.1 [dev] transformer, BF16 weights | 24 GB, plus text encoders and VAE | ✓ |
| FLUX.1 [dev] LoRA training, BF16 base | ≈ 28.8 GB | ✓ |
| SDXL 1.0 base, FP16 weights | 7 GB | ✓ |
| Llama 3.1 70B inference, 4-bit | ≈ 42 GB | ✗ |
| Llama 3.1 8B full fine-tune | ≈ 128 to 144 GB before activations | ✗ |
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.
Popular
Day
24 GPU-hours
$10.08
$0.42/hour
- One day of continuous running
- A fine-tuning run or a batch job
- Unused hours never expire
Week
168 GPU-hours
$70.56
$0.42/hour
- One week of continuous running
- Training runs and evaluations
- Unused hours never expire
Month
720 GPU-hours
$302.40
$0.42/hour
- 30 days of continuous running
- An inference server running all month
- Unused hours never expire
In every plan: Whole GPUs, never shared · Root SSH, CUDA and PyTorch ready · Billed per minute · Unused hours never expire
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.
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 with no ID and no card. Salad, Spheron and RunPod also accept crypto, but RunPod asks for KYC before the first crypto payment.
Sources (12)
- SaladCloud price changes, November 2026 blog.salad.com/saladcloud-price-changes-november-2026 — Salad (read 2026-10-29)
- SaladCloud — Billing docs.salad.com/general/explanation/billing — Salad (read 2026-10-29)
- Spheron — Pricing spheron.network/pricing — Spheron (read 2026-10-29)
- Spheron — Billing — Spheron (read 2026-10-29)
- Runpod — Pricing runpod.io/pricing — RunPod (read 2026-10-29)
- Runpod — Billing information docs.runpod.io/references/billing-information — RunPod (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)
- Shadeform — Instance types API api.shadeform.ai/v1/instances/types — Shadeform (read 2026-10-29)
- NVIDIA GeForce RTX 5090 — NVIDIA (read 2026-10-29)
- NVIDIA RTX Blackwell GPU architecture whitepaper — NVIDIA (read 2026-10-29)
- Transformer Math 101 — EleutherAI (read 2026-10-29)