Cloud GPU prices, by VRAM
Not enough VRAM at home? Renting a cloud GPU by the hour is the quickest way to run a bigger model. Pick a model below to see the cheapest setup that holds it, then check the full price list.
Cheapest cloud setup for a model
Pick a model. We check 1, 2, 4 and 8 of every GPU below and list the cheapest setups that hold it.
Fit rule: weights + KV cache under 90% of the combined VRAM, split evenly across the GPUs (llama.cpp, vLLM and Ollama can do this). Prices checked 2026-10-23.
All on-demand prices
49 on-demand offers, sorted by VRAM. Tap a heading to re-sort. Prices are US dollars per GPU per hour, before tax, copied from each provider's own pricing page.
| Tesla V100 | 16 GB | $0.798-GPU machines only | 4.94¢ | LambdaOn-demand instance |
|---|---|---|---|---|
| RTX A5000 | 24 GB | $0.16 | 0.67¢ | RunpodCommunity Cloud |
| RTX 3090 | 24 GB | $0.22 | 0.92¢ | RunpodCommunity Cloud |
| RTX A5000 | 24 GB | $0.27 | 1.13¢ | RunpodSecure Cloud |
| RTX 4090 | 24 GB | $0.34 | 1.42¢ | RunpodCommunity Cloud |
| L4 | 24 GB | $0.44 | 1.83¢ | RunpodCommunity Cloud |
| RTX 3090 | 24 GB | $0.50 | 2.08¢ | RunpodSecure Cloud |
| L4 | 24 GB | $0.59 | 2.46¢ | RunpodSecure Cloud |
| Quadro RTX 6000 | 24 GB | $0.69 | 2.88¢ | LambdaOn-demand instance |
| RTX 4090 | 24 GB | $0.89 | 3.71¢ | RunpodSecure Cloud |
| A10 | 24 GB | $1.29 | 5.38¢ | LambdaOn-demand instance |
| RTX 5090 | 32 GB | $0.69 | 2.16¢ | RunpodCommunity Cloud |
| RTX 5090 | 32 GB | $1.19 | 3.72¢ | RunpodSecure Cloud |
| A100 PCIe | 40 GB | $1.99 | 4.98¢ | LambdaOn-demand instance |
| A100 SXM | 40 GB | $1.99 | 4.98¢ | LambdaOn-demand instance |
| RTX A6000 | 48 GB | $0.33 | 0.69¢ | RunpodCommunity Cloud |
| A40 | 48 GB | $0.35 | 0.73¢ | RunpodCommunity Cloud |
| RTX A6000 | 48 GB | $0.59 | 1.23¢ | RunpodSecure Cloud |
| A40 | 48 GB | $0.59 | 1.23¢ | RunpodSecure Cloud |
| L40 | 48 GB | $0.69 | 1.44¢ | RunpodCommunity Cloud |
| RTX 6000 Ada | 48 GB | $0.74 | 1.54¢ | RunpodCommunity Cloud |
| L40S | 48 GB | $0.79 | 1.65¢ | RunpodCommunity Cloud |
| L40 | 48 GB | $0.82 | 1.71¢ | RunpodSecure Cloud |
| RTX 6000 Ada | 48 GB | $0.99 | 2.06¢ | RunpodSecure Cloud |
| L40S | 48 GB | $1.09 | 2.27¢ | RunpodSecure Cloud |
| RTX A6000 | 48 GB | $1.09 | 2.27¢ | LambdaOn-demand instance |
| A100 PCIe | 80 GB | $1.19 | 1.49¢ | RunpodCommunity Cloud |
| A100 SXM | 80 GB | $1.39 | 1.74¢ | RunpodCommunity Cloud |
| A100 PCIe | 80 GB | $1.79 | 2.24¢ | RunpodSecure Cloud |
| A100 SXM | 80 GB | $1.79 | 2.24¢ | RunpodSecure Cloud |
| H100 PCIe | 80 GB | $1.99 | 2.49¢ | RunpodCommunity Cloud |
| H100 SXM | 80 GB | $2.69 | 3.36¢ | RunpodCommunity Cloud |
| A100 SXM | 80 GB | $2.798-GPU machines only | 3.49¢ | LambdaOn-demand instance |
| H100 PCIe | 80 GB | $2.89 | 3.61¢ | RunpodSecure Cloud |
| H100 PCIe | 80 GB | $3.29 | 4.11¢ | LambdaOn-demand instance |
| H100 SXM | 80 GB | $3.99 | 4.99¢ | RunpodSecure Cloud |
| H100 SXM | 80 GB | from $3.991 GPU: $4.292 GPUs: $4.194 GPUs: $4.098 GPUs: $3.99 | 4.99¢ | LambdaOn-demand instance |
| H100 NVL | 94 GB | $2.59 | 2.76¢ | RunpodCommunity Cloud |
| H100 NVL | 94 GB | $3.19 | 3.39¢ | RunpodSecure Cloud |
| RTX Pro 6000 | 96 GB | $1.69 | 1.76¢ | RunpodCommunity Cloud |
| GH200 | 96 GB | $2.29 | 2.39¢ | LambdaOn-demand instance |
| RTX Pro 6000 | 96 GB | $2.49 | 2.59¢ | RunpodSecure Cloud |
| H200 | 141 GB | $3.59 | 2.55¢ | RunpodCommunity Cloud |
| H200 | 141 GB | $5.29 | 3.75¢ | RunpodSecure Cloud |
| B200 | 180 GB | $5.98 | 3.32¢ | RunpodCommunity Cloud |
| B200 SXM6 | 180 GB | from $6.691 GPU: $6.992 GPUs: $6.894 GPUs: $6.798 GPUs: $6.69 | 3.72¢ | LambdaOn-demand instance |
| B200 | 180 GB | $7.99 | 4.44¢ | RunpodSecure Cloud |
| B300 | 288 GB | $6.94 | 2.41¢ | RunpodCommunity Cloud |
| B300 | 288 GB | $8.99 | 3.12¢ | RunpodSecure Cloud |
Checked 2026-10-23. GPU prices change often and stock varies by region, so confirm on the provider's page before you rent. Where Lambda lists different per-GPU rates for 1, 2, 4 or 8 GPUs we show each one. Runpod lists one per-GPU rate. These are plain links, not affiliate links.
How to read this
- Per GPU, per hour. A setup with 2 GPUs costs twice the per-GPU rate. Where Lambda lists a different rate for bigger machines, we use that rate.
- VRAM decides what fits. The setup finder uses the same 90% rule as our VRAM calculator: weights plus KV cache must stay under 90% of the combined VRAM.
- Only two providers so far. We only list prices we could read on the provider's own pricing page. More providers will be added as we verify them.
- No affiliate links. None of these links earn us anything.
Download the data: models (CSV) · models (JSON) · GPU prices (JSON). Every row carries its source links and checked date.