Unix & Linux Stack Exchange is a question and answer site for users of Linux, FreeBSD and other Un*x-like operating systems. It only takes a minute to sign up.
Sign up to join this community
I have a nvidia gpu which I use for passthrough to a =Windows VM. However, sometime I would like to have it accesible in my linux host (while the Windows VM is down), for example to be able to use for video encoding.
Is there anyway to unload the vfio driver and load the nvidia one on demand?
To my knowledge only more expensive cards made specifically for virtualization support this properly. So be prepared for very strange problems including complete freezes as the card freaks out in a weird state. (Even if you find a way.)
And used to forbid Nvidia even forbids it on other cards. Which is why last time I checked you needed some hacks to hide that. Nividia only recently changed their stance on this:
https://nvidia.custhelp.com/app/answers/detail/a_id/5173/~/geforce-gpu-passthrough-for-windows-virtual-machine-%28beta%29
"Enabled" does not equal "Supported"
I currently have a pass through setup running and have been using it regularly for a few years now. And my experience tells me: You are in for a world of pain. Those other cards cost a lot more. So Nvidia isn't exactly incentivized to support this.
That said, I hope you succeed. I just wanted to tell you: Don't hold your breath you are working against the odds. Be prepared for heavy tinkering. Good luck!
Update:
Mind sharing your setup?
Mainboard and GPU would be most interesting.
modprobe -r vfio-pci && modprobe nvidianot work?vfio-pci.ids=10de:1c03on your kernel cmd line), and IIRC you can't change that at run time. Have you considered doing your video encoding in a VM? either your existing VM or maybe a new Linux VM (which will probably be easier to script the transcode than windows)...just don't run both windows and linux vm using the gpu at the same time.