From sign-up to SSH. In a few steps.
Add an SSH key, rent a machine, connect as root and get to work. Everything below is plain shell on your own machine.
1. Add an SSH key. Then rent a machine.
Create a key if you don’t have one, then paste the public key into the SSH keys page of the console. It is installed on every machine you rent.
To rent, pick a GPU model, the GPU count, an image and the number of hours. Check the quote, pay from your balance, and the machine is usually ready in about a minute.
# Create a key if you don't have one
ssh-keygen -t ed25519
# Print the public key, then paste it into
# the console under SSH keys
cat ~/.ssh/id_ed25519.pub2. Connect as root. Check your GPUs.
The instance page shows the host and port for your machine. Connect as root with your key.
Run nvidia-smi to see the GPUs attached to the machine. Use tmux so long jobs keep running if your connection drops.
# Host and port are on the instance page
ssh root@<host> -p <port>
# List the GPUs on this machine
nvidia-smi
# Keep a session running after you disconnect
tmux new -s work3. Keep your work on /data. Run what you like.
If you added a data disk, it is mounted at /data (ext4) and kept across restarts of that machine.
Docker, Python 3 with pip, git, rsync and gcc are already installed, so you can clone a repository and start straight away.
# The optional data disk
df -h /data
# Clone your code onto it
cd /data && git clone <your-repo-url>
# Docker is installed too
docker ps4. Copy results out. Release when you’re done.
Releasing deletes the machine and its data disk unless you chose to keep the disks, so copy out anything you want to keep. Then release from the console and confirm the machine ID.
Unused whole hours of compute and data disk go back to your balance automatically.
# Run this on your own computer
rsync -avP -e "ssh -p <port>" \
root@<host>:/data/results/ ./results/Questions, answered.
Which image should I pick?
There are two: Ubuntu 22.04 with CUDA 12.4, and Ubuntu 24.04 with CUDA 13.0. The RTX 5090 requires Ubuntu 24.04 with CUDA 13.0.
What comes installed?
Both images include the NVIDIA driver, Docker, Python 3 and pip, git, tmux, htop, nvtop, jq, rsync and gcc.
Can I restart a machine?
Yes, from the console. The data disk is kept across restarts of the same machine.
What time zone and locale do machines use?
UTC and an English locale, on every machine.
Something isn’t working. Who do I ask?
Email [email protected] with your machine ID and what you tried.