I kinda forgot what this thread was about from the original post and on, but I just want to report that I have an RTX4070 in my laptop, and I can pass it through some debian 12 qubes as PCI device. And nvidia-driver (installed from bookworm-backports, or testing) works with it. I can run ffmpeg with hardware encoder options, or run video games insside Flatpak Lutris – both make use of my GPU. Oh, I can also make ollama.com tool use my GPU passthrough device for running some large language models locally.
I don’t have the knowledge or need to really discern whether the performance I am getting from my GPU via passthrough is good, or whether I am leaving much of the hardware performance on the table; all I can say is I am currently content with how it works.