Support & Issues
#@ Plug & Play This is not a "Plug & Play" solution. You'll very probably need to build an image
If you have multiple nodes able to run this, but not the same configuration (for example one node with AMD GPU and another with NVIDIA), you'll not be able to use the same image in the system. You'll need to set node affinity rules to make sure the image runs in a supported node.
That being said, I'm assuming most of you will have 1 dedicated node, or otherwise repeatable configuration if you're running a GPU cluster.
For some reason, the AMD GPU I am using to run Plasma was not
being initialized properly when running the container. It would
not display GPU apps, but only CPU-rendered apps (like default
window manager stuff, xfishtank...).
Running glmark2 for a little while "primes" the GPU and enables
all other apps to run.
For this reason there is a priming service in Plasma, and active by default for AMD cards.
Feel free to test without it (set the GPU_PRIME env value to empty)
and report your findings!
When using multiple GPUs (notably a normal setup with an iGPU and a dGPU), it is possible X and the graphics drivers can get confused.
To fix this the best solution would be to add the PCI ID to the X config
and/or force the /dev/dri card
For now the resolution is locked at 1920 x 1080, @ 60Hz.
There is a solution to this using cvt and adding the modeline
manually to the X configuration.
The Steam-headless project also uses a special xrandr configuration
for when clients connect through Sunshine, dynamically changing the resolution.