If I have a Ubuntu desktop build that would hypothetically have a NVIDIA GPU (3080TI) in addition to an AMD card (say that Mi25), is there a way to specify a graphics card to a particular process/purpose? Just using these cards as examples. I suppose another case would be an Intel iGPU and some non-intel PCIe card.
Plain examples
Card A: Gaming
Card B Stable Diffusion
or like
Card A: Handbrake
Card B: either from example 1.
Let me know your thoughts. And I appreciate the input as always.
WWED
EDIT: I’m sure pass-through to a VM would work but I imagine somewhere there would be a performance hit.
Multiseat would do this, but it’s a bit of a pain. Basically, each GPU acts as it’s own independent desktop. They share the root filesystem, but can’t really talk to eachother otherwise.
Not sure about wayland support. It was easy enough with xorg back when I was using it with a couple nvidia gpus, but something happened that broke it and I’m not sure the feature currently works properly in the kernel right now.
This was for display output though. I think, for command line stuff, you should just be able to specify a GPU and have it use that GPU for whatever. But, for multiseat, you could have a main seat for gaming/transcoding video via nvenc, and a second seat for workstation stuff, with it’s own dedicated monitor, mouse/KB, etc.