ubuntu-ssh-cuda — bare Ubuntu + CUDA + SSH
A minimal Ubuntu 22.04 + CUDA 12.4 image with just SSH and the essentials. Use this when you want full control over the stack and you’ll install everything yourself.
Image tag
docker.io/manvarharsh/ubuntu-ssh-cuda:cuda12
What’s in this image
- Base:
nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04 - Python 3.10 (conda) — minimal
- Hugging Face Hub CLI (handy for downloading models)
- OpenSSH server
- No JupyterHub, no Python ML libs preinstalled
Default ports
| Port | Service |
|---|---|
| 22 | SSH |
Use cases
- Bring-your-own-stack environments
- Reproducing a colleague’s exact
pip installsetup - Testing custom CUDA code without framework interference
- Lightweight remote dev box for GPU work
Environment variables
| Variable | Description |
|---|---|
ENABLE_SSH | Enable SSH server (typically true for this image) |
SSH_PUBLIC_KEY | Public key for SSH |
Persistence
Mount at /data. Install your stack into /data/venv/ (or use pip install --user) so it survives restarts.
See also
- Conventions and shared environment variables
- jupyterlab-gpu — if you want JupyterLab pre-wired
- Accessing output files in the browser