MATE Desktop container with GLX support for NVIDIA GPUs by spawning its own X Server and noVNC WebSocket interface instead of using the host X server. Does not require /tmp/.X11-unix host sockets or any non-conventional/dangerous host setup.
Use docker-nvidia-egl-desktop for a more stable MATE Desktop container that directly accesses NVIDIA GPUs without using an X Server.
Change the NVIDIA GPU driver version to be same as the host and build your own Dockerfile. Change bootstrap.sh if you are using a headless GPU like Tesla. Corresponding container toolkit on the host for allocating GPUs should also be set up.
Connect to the spawned noVNC WebSocket instance with a browser in port 5901, no VNC client required.
Note: Requires access to at least one /dev/ttyX device. Check out k8s-hostdev-plugin for provisioning this in Kubernetes clusters without privileged access.
For Docker this configuration is tested to work but the container will have potentially dangerous privileged access:
docker run --gpus 1 --privileged -it -e SIZEW=1920 -e SIZEH=1080 -e VNCPASS=vncpasswd -p 5901:5901 ehfd/nvidia-glx-desktop:latest
The below may also work without privileged access but is untested and may be buggy:
docker run --gpus 1 --device=/dev/tty0:rw -it -e SIZEW=1920 -e SIZEH=1080 -e VNCPASS=vncpasswd -p 5901:5901 ehfd/nvidia-glx-desktop:latest