Frequently Asked Questions

1. What is gpu.st?

gpu.st is a service that gives you access to a GPU instance without any hassle or ceremony. Every instance has a persistent root filesystem, one or more GPUs attached, and an open port 80 for easily using and sharing web demos.

2. Why should I care?

Not everyone has a machine with a powerful GPU at home; and not everyone wants to go through the hassle of spinning up a machine on a cloud provider, picking the right container image, configuring 50 other options, waiting for it to start, and then getting the information needed to log in over SSH.

Running inference with vLLM? Messing around in Jupyter notebooks? Sharing a demo? Doing a quick training run or testing a kernel? It's only a single ssh gpu.st new away.

3. What GPUs are available?

The list of available GPUs is available through the gpus command and here. If you're logged in, it'll only show you the GPUs available on your current plan by default; note that the free plan only has a limited number of GPU types available!

4. What's this about a web server?

We operate a reverse proxy, turning your-instance.mygpu.ws into an HTTP request to port 80 on your instance. We have built-in but optional authentication, keeping your work safe and secure, while giving you the option to share it with the world later.

5. Is it really that easy?

Yes. By default, all instances are created from an image of GPUbuntu, our mini-distribution of Ubuntu which includes various tools and libraries from the CUDA toolkit to the latest PyTorch.

We also support most Debian, Ubuntu, and Alpine-based container images.

6. How do I move files around?

You can use sftp, scp, and rsync to transfer files to and from your instance just like any other Linux machine. Of course, you're free to install other tools such as rclone and Syncthing at your leisure.

7. I didn't receive the verification email!

If you don't receive your verification email after five minutes or so, try whitelisting the sender support@gpu.st. For many providers, sending an email to support@gpu.st with the subject "whitelist" also works. After you've done so, try requesting another verification email.