I can only answer for NVIDIA, as that was the platform we developed on, but it should be possible to encapsulate the X11 server within the container without starting an X server on the host machine, and then plug in to the monitor you designate.

I can answer further if you are satisfied with the directions found within https://github.com/selkies-project/docker-nvidia-glx-desktop. It has a lot of effort to NOT require a monitor, but it should also work with one.

It's possible to use tini, supervisord, dumb-init, and other init 1 services for containers. Bit of a heavy workaround to make it work with desktop environments, but works.

Selkies should be generally acceptable for gaming, provided that you go through a bit of the WebRTC setup.

GNOME forces OpenGL 3D compositing with no way to turn off, which is very uncomfortable without a GPU. Every other X11 desktop environment allows to turn them off, which is more tolerable when using llvmpipe, the software CPU Mesa OpenGL implementation.

Maintainer here! Thank you for the post.

While our focus is for delivering high-performance remote 3D graphical workloads for vanilla Kubernetes and HPC clusters in general, our project should be trivial for anyone to also deploy in their X11 desktop, especially with NVIDIA.