Live GPU prices
Every GPU type below is rentable as a pod with per-minute billing. Prices are read from the live catalog when the page loads.
| GPU | VRAM | Secure cloud $/hr | Community cloud $/hr | Max GPUs per pod |
|---|---|---|---|---|
| B300 | 288 GB | $9.71 | $7.50 | 8 |
| MI300X | 192 GB | $2.58 | — | 8 |
| B200 | 180 GB | $8.63 | $6.46 | 8 |
| H200 NVL | 143 GB | $4.09 | — | 8 |
| H200 SXM | 141 GB | $5.71 | $3.88 | 8 |
| RTX PRO 6000 | 96 GB | $2.69 | $1.83 | 9 |
| RTX PRO 6000 MaxQ | 96 GB | — | $1.77 | 2 |
| RTX PRO 6000 WK | 96 GB | $2.80 | $1.83 | 8 |
| H100 NVL | 94 GB | $3.45 | $2.80 | 10 |
| A100 PCIe | 80 GB | $1.93 | $1.29 | 8 |
| A100 SXM | 80 GB | $1.93 | $1.50 | 8 |
| H100 PCIe | 80 GB | $3.12 | $2.15 | 8 |
| H100 SXM | 80 GB | $4.31 | $2.91 | 8 |
| A40 | 48 GB | $0.64 | — | 10 |
| L40 | 48 GB | $0.89 | $0.75 | 10 |
| L40S | 48 GB | $1.18 | $0.85 | 8 |
| PRO 6000 MIG 48GB | 48 GB | $1.39 | — | 16 |
| RTX 6000 Ada | 48 GB | $1.07 | $0.80 | 8 |
| RTX A6000 | 48 GB | $0.64 | $0.36 | 10 |
| RTX PRO 5000 | 48 GB | $1.04 | $0.89 | 8 |
| A100 SXM 40GB | 40 GB | — | $1.08 | 2 |
| RTX 5000 Ada | 32 GB | — | $0.53 | 4 |
| RTX 5090 | 32 GB | $1.29 | $0.75 | 8 |
| RTX PRO 4500 | 32 GB | $0.78 | — | 8 |
| RTX PRO 4500 SE | 32 GB | $0.78 | — | 8 |
| L4 | 24 GB | $0.64 | — | 9 |
| PRO 6000 MIG 24GB | 24 GB | $0.75 | — | 32 |
| RTX 3090 | 24 GB | $0.54 | $0.24 | 8 |
| RTX 3090 Ti | 24 GB | — | $0.29 | 4 |
| RTX 4090 | 24 GB | $0.96 | $0.37 | 8 |
| RTX A5000 | 24 GB | $0.29 | $0.17 | 10 |
| RTX PRO 4000 | 24 GB | $0.62 | — | 4 |
| RTX 4000 Ada | 20 GB | $0.30 | $0.22 | 8 |
| RTX 4000 Ada SFF | 20 GB | — | $0.20 | 7 |
| RTX A4500 | 20 GB | $0.27 | $0.21 | 4 |
| RTX 2000 Ada | 16 GB | $0.26 | — | 8 |
| RTX 4080 | 16 GB | — | $0.29 | 2 |
| RTX 4080 SUPER | 16 GB | — | $0.30 | 4 |
| RTX 5080 | 16 GB | — | $0.42 | 8 |
| RTX A4000 | 16 GB | $0.27 | $0.18 | 8 |
| Tesla V100 | 16 GB | — | $0.21 | 8 |
| V100 SXM2 | 16 GB | — | $0.25 | 8 |
| RTX 3080 Ti | 12 GB | — | $0.20 | 4 |
| RTX 4070 Ti | 12 GB | — | $0.21 | 5 |
| RTX 3080 | 10 GB | — | $0.18 | 2 |
| RTX 3070 | 8 GB | — | $0.14 | 6 |
| RTX A2000 | 6 GB | — | $0.13 | 4 |
Which GPU should you rent?
- RTX 4090 (24 GB) — the cheapest way to run 7B–14B models, Stable Diffusion and LoRA fine-tunes. Currently $0.96 per hour.
- A100 (40/80 GB) — a reliable workhorse for 30B–70B inference with quantisation and for mid-size training runs. Currently $1.08 per hour.
- H100 (80 GB) — the default for serious training and high-throughput inference with FP8 support. Currently $3.12 per hour.
- H200 (141 GB) — more memory per GPU, so large models fit on fewer cards and long-context inference has room for the KV cache.
A practical rule: model weights in FP16 need about 2 GB of VRAM per billion parameters, plus headroom for the KV cache. A 70B model needs roughly 140 GB in FP16, or about 40 GB at 4-bit.
How billing works
Pods bill per minute while they exist. Stopping a pod stops GPU charges, but attached persistent volumes keep billing for storage until you delete them. The configurator shows the hourly and monthly price for GPU, region, template and disk before anything is provisioned.
GPU cloud or a serverless API?
If you only need to call a popular model, the pay-per-token API is usually cheaper and has nothing to manage: you pay only for tokens. Rent a GPU when you need custom or fine-tuned weights, sustained high throughput where a dedicated card beats per-token pricing, training, or full control of the runtime (vLLM, TGI, ComfyUI, your own container). Both products use the same FastInfra account and prepaid balance.
Deploy in three steps
- Pick a GPU type and region in the configurator.
- Choose a container template and size the persistent volume.
- Deploy from the dashboard and connect over SSH or Jupyter.