Docker Image :: Ubuntu KDE Plasma Desktop with PyTorch, CUDA & VNC

Ideal for LLM & Deep Learning remote work.

Screenshot

Content

Base

• PyTorch (base image including Ubuntu & CUDA)
• KDE Plasma Desktop
• VNC Server (port 5900)

Main Applications

• Firefox (ESR)
• Kitty (terminal)
• Sublime Text
• Visual Studio Code (must be run with --no-sandbox flag)

Nice to Have

• Fira Code (monospace font)
• zsh (with oh-my-zsh)

Tags

Pytorch tag as per their official documentation:

• PyTorch 23.09-py3
  • latest

Environment Variables

• VNC_PASSWORD: Password for VNC connection. Default: password.
• VNC_SCREEN_SIZE: Screen size. Default: 1920x1080.

Usage

To connect to the container's desktop, you can use any VNC client.

[!NOTE]
You will have a wide list of errors during startup but you can ignore them. Enulating a full X server in Docker is not an easy task.

Your server will be ready once you see this line in the logs:

[...]
cuda-desktop-server  | Plasma Shell startup completed
[...]

Deployment

RunPod

[!NOTE]
You can create and attach a Storage Network Volume to your RunPod pod to automatically sync and persist your /home/ubuntu directory between deployments.

Backup Volume in Deployment

When deploying your container, you can also attach a shared storage as a volume for backup.

Simply mount your desired backup directory to /volumes/home_ubuntu_backup in your Docker run command or Docker composer file. It will be automatically retored (on subsequent boots) and then synced each minute.

Thanks

• @ms-jpq for ms-jpq/kde-in-docker
• Selkies for selkies-project/docker-nvidia-glx-desktop