GPUbuntu
GPUbuntu is gpu.st's special Ubuntu distribution that comes preloaded with tools and libraries for AI/ML research and scientific computing. It is the default image used for new instances on the platform.
GPUbuntu is a standard Ubuntu image, and additional packages can be installed
using apt as well using the gpust-install command (described below).
Pre-installed software
The default installation includes the Caddy web server, an appropriate version
of the CUDA toolkit, Python 3.14.2, JupyterLab 4.5.2, and Pytorch 2.9.1. You
can specify additional software to preinstall by setting the
GPUST_AUTO_INSTALL environment variable when you create your instance:
new --env GPUST_AUTO_INSTALL=pytorch,cuda-toolkit,jupyter.
A Python virtual environment named default is created for you at
/root/default.
First boot
When you start your instance for the first time, GPUbuntu will perform some housekeeping by installing the requested software packages and configuring built-in services. This may take a few minutes, but subsequent starts should be almost instantaneous.
Web services
The web server is configured to listen on port 80, making it accessible at
https://your-instance.mygpu.ws. After authenticating, you'll be presented
with a list of enabled, running services you can access, including JupyterLab.
Supervisor
Supervisor is used for managing services on GPUbuntu, and you can control
running services through supervisorctl. You can configure your own services
by placing configuration files in /root/etc/supervisord/.
Other packages
GPUbuntu comes with tmux, jq, netcat, C/C++ development tools, and other
packages so you don't have to install them later.
Commands
gpust-status
The gpust-status command prints the current status of your instance,
including kernel version, supported CUDA version, installation status of
components such as PyTorch and CUDA, and running status of services such as
Caddy and JupyterLab.
gpust-install
The gpust-install command allows you to install additional software that
might be useful. You can run gpust-install to see a list of software that
you can install, or refer to this (potentially outdated) list below:
AI/ML Libraries
- pytorch (2.9.1)
- jax (0.8.2)
- tensorflow (2.20.0)
- cuda-toolkit (appropriate version autodetected)
- huggingface
- ml-utils: NumPy, etc.
Development Environments
- jupyter (4.5.2)
- code-server: VSCode
Coding and AI Agents (coding-agents)
- claude-code
- codex-cli
- opencode
Network tools
- socat
- cloudflared