GPU capacity with clear terms.
Compare hardware, location, trust state and hourly price before you reserve. Prices and limits are shown before funding; billing follows accepted meter evidence, not estimates.
Current offers, compared
| Machine | VRAM per GPU | Region | Trust state | Total rate | Details |
|---|---|---|---|---|---|
| 2 × NVIDIA GeForce RTX 4090 | 24 GiB | eu-central | Verified | $0.70/h ≈ $0.35/GPU·h | Details |
| 1 × AMD Instinct MI210 | 64 GiB | us-west | Verified | $0.45/h | Details |
How the price works: the total machine rate covers the whole machine — all GPUs, included disk and egress, and the platform's 10% share. The per-GPU figure is the same rate divided for comparison; you always rent the full machine.
Rent a GPU
- Compare offers and pick a template that meets the supported profile.
- Set a bounded budget and duration; review the immutable quote.
- Fund, then connect with copyable SSH local forwarding once the machine verifies ready.
- Stop any time; results stay readable for 30 minutes, then are deleted.
Put available GPUs to work
- Run the non-mutating compatibility check before changing anything.
- See exact install changes, enroll with a scoped one-use token.
- Declare availability windows and a rate; publish after tests pass.
- Follow each job from reservation through settlement; fee is a flat 10%.
Explicit trust state
Detection, runtime availability, qualification and rentability are shown separately. Verification is never implied by a badge alone.
Evidence-backed billing
Charges follow control-plane accepted meter evidence and the durable journal. Missing evidence stays unavailable — never converted to a success.
One clear lifecycle
Reserved, funded, running, stopping, settled. Stop-requested and stop-confirmed are different states, and you can always see which one you are in.