GPU Cloud · Billed per minute
Price your GPU pod before you buy
Pick a GPU, size the disks, and see exactly what it costs per minute, per day, and per month. Nothing is provisioned and nothing is charged until you deploy — configure freely.
1 · Choose a GPU
2 · Size your pod
Nothing is reserved while you configureHolds your image, CUDA cache, and scratch files. Ephemeral — wiped when the pod stops.
Mounted at /workspace and survives stops. Datasets and checkpoints belong here.
How long this configuration can run on that much credit, assuming it stays running the whole time.
What fits in — of VRAM
Rule-of-thumb weight sizing- FP16 / BF16
- — 2 bytes per parameter
- 8-bit
- — 1 byte per parameter
- 4-bit
- — 0.5 bytes per parameter
Typically used for: —
Estimates reserve 20% of VRAM for the KV cache and activations. Real headroom depends on context length, batch size, and your serving stack — treat these as a starting point, not a guarantee.
—
- 1 hour
- —
- 1 day
- —
- 1 week
- —
- 1 month
- —
Minimum credit to start this pod: — (1h of runtime). Compute billing stops the moment you stop the pod.
Billed by the minute
Charges are drawn from your prepaid credits every minute the pod is provisioning or running. Stop it and compute charges end instantly — no minimum term, no reservation fees.
Secure vs value tier
Secure runs in vetted datacenters with the highest reliability and availability. Value uses lower-cost community hosts — same GPU model, cheaper rate, more variable capacity.
Storage that persists
A persistent volume survives stops and restarts. A stopped pod only pays the volume storage rate ($0.216/GB/month) — never for the GPU.
Transparent pricing
Every rate shown includes our 8% platform markup on underlying compute and storage — there is nothing added at checkout. Up to 2 active pods per account by default.
What you get on every pod
Your container, your rules
Any public Docker image — PyTorch, TensorFlow, vLLM, ComfyUI, Ollama, or your own build. Set environment variables and expose up to 10 ports.
Direct access
Enable SSH and JupyterLab at deploy time. TCP ports get a public IP and port you connect to directly; HTTP ports are proxied for you.
One wallet, one dashboard
GPU pods draw from the same FastInfra credits as API inference. Live usage, per-minute charges, and runtime totals appear on each pod's detail page.