Rent your first GPU in 5 minutes

From an empty account to a running container with a terminal open: sign up, top up, pick a machine, rent, connect, stop.

You rent a real person's graphics card. The container starts on their PC, you reach it through the FairGPU relay, and you pay per second from a prepaid wallet. Nothing can run up a bill you did not fund.

1. Create an account and top up

1. Register with email and password, or with Google / GitHub / Apple. Verify the email (the link is in your inbox).
2. Open Wallet → Top up. The minimum is small; $5 is plenty for a first hour on most cards. Credits do not expire and are only ever spent on rentals.
3. You do not need an SSH key to start: the browser terminal works without one. Add a key later for ssh, scp and VS Code (see Connect).

2. Pick a machine

Open the marketplace. Every card is one PC, priced by its owner. Things worth a look:

Click a card for the full listing: photos, calendar, specs, reviews, and the What happens after you book strip that spells out billing start, the policy and what you need.

3. Rent

Click Rent (or Request to book on request-mode listings: the host approves within 24 h and nothing is charged until then).

1. Mode. Interactive = a machine you log into (terminal, SSH, Jupyter, web UIs). Job = a command that runs and streams its logs; tick Keep the container running to run more commands afterwards.
2. Template. Official templates set the image, ports and env for you. Custom image takes any public Docker image. Fill in required settings (for example MODEL for vLLM); secret-looking values can be stored once as secrets and referenced by name.
3. SSH keys (interactive): tick the keys to inject, or none. Maximum duration: a hard cost cap; you can extend while it runs. Auto-stop when idle is on by default and free. Checkpoints: on servers with FairGPU cloud storage you can snapshot /workspace every 15, 30 or 60 minutes.
4. Storage (when the host offers it): a persistent /workspace that survives stops, priced per GB-month by the host.
5. Review: the quote shows the host's share and the platform fee, the interruption policy and the refund rule that applies. Start rental.

4. Watch it start, then connect

The rental page opens. The stepper goes Starting → Running; the Activity timeline shows the image pull layer by layer. You are billed from Running, not before.

The Getting started card walks you through the next steps with live ticks:

1. Container ready.
2. Connect – choose one:

• Terminal (browser, no key): click Terminal on the rental page. Best way to start.
• SSH: ssh -p <port> root@<relay> from the Connect dialog, which also shows root@host -p port, the SFTP command and your host-key fingerprint.
• Jupyter: one click when the image exposes port 8888.
• VS Code: Remote-SSH one-liner and a ready ~/.ssh/config block.

1. Put your data in /workspace (commands for your OS are in the card and in Move files).
2. Run your work: the template's own guide appears here (where to put models, which port to open, how to verify the GPU).
3. Save results: download, checkpoint to FairGPU storage, or push to your own bucket.
4. Stop to stop billing.

First thing to type in the terminal:

nvidia-smi
python -c "import torch; print(torch.cuda.is_available(), torch.cuda.get_device_name(0))"

5. Stop

Click Stop on the rental page (or let idle auto-stop do it). Compute billing ends within seconds. If you kept a workspace, it stays on the host and keeps billing at the host's storage rate until you delete it under Workspaces; otherwise the container and everything outside a checkpoint is gone.

Leave a review: it is how good hosts rise.

Next

• Connect: terminal, SSH, Jupyter, VS Code
• Move files in and out
• Keep installs between runs
• Billing and refunds