Docs
Quickstart
From access request to a shell on a datacenter GPU. Request access, book a machine for 1 to 24 hours, connect over SSH, end the session.
Request access, book a datacenter GPU, open a shell. About ten minutes from invitation to nvidia-smi.
1. Request access
Fill in the access form: email, company, and what you will run. Invitations go out by email.
2. Accept your invitation
The invitation email carries a sign-up link and, when granted, compute credit for your organization.
- Open the link. It works once and expires after 7 days.
- Create your account with the invited address and accept the terms.
- Confirm your email from the verification message.
Your credit appears under Account → Credit.
3. Pick a machine
Open Catalog → Data-center capacity. Each offer lists:
- GPU model and count, VRAM per GPU
- vCPU, RAM, disk
- Region label (for example
CANADA-1) - Stock, refreshed every few minutes
- Hourly rate
Offers marked interruptible run on spot capacity. See Sessions and limits.
4. Book
- Open Details on an offer.
- Set Duration: 1 to 24 hours.
- Select Book this configuration.
The cost of the full duration is held from your credit. You pay only for the minutes used. See Billing.
The booked window starts now. Start the machine right away.
5. Start the machine
On the reservation page:
- Paste your SSH public key (the content of your
.pubfile). - Enter your public IP address: the address you will connect from.
- Select Create my machine.
Provisioning takes about 2 minutes. The page updates by itself.
6. Connect
When the machine is ready, the page shows the command:
ssh ubuntu@203.0.113.42
Check the GPU:
nvidia-smi
The machine runs Ubuntu 24.04 with the NVIDIA driver and CUDA 12.8. Details in Connect over SSH.
7. End the session
Copy your results off the machine first:
rsync -avz ubuntu@203.0.113.42:~/results/ ./results/
Then select End session. Billing stops immediately. The machine and its disk are deleted.
The session also ends on its own when the booked time is up.
Next
- Connect over SSH
- Sessions and limits
- Recipes: vLLM, ComfyUI, LoRA fine-tuning, Blender