Set passthrough GPU as a primary graphics card for a VM

kubuntu-newbie

@miltador have you ever been able to figure this out? I'm having the same problem with plasmashell trying to use virtual gpu instead of nvidia

miltador

@kubuntu-newbie nah, I gave up on this. I think it's generally not suited for such use cases.
PCI passthrough works well in XCP-ng, GPU passthrough - nope unless you have special hardware

kubuntu-newbie

@miltador

I was able to get it working with kubuntu from 22.04 through 25.10, running X11 plasma session just by passing through PCI and running "nvidia-xconfig --enable-all-gpus --separate-x-screens to create xorg.conf" in the VM.

It used to work with Wayland as well on XCP-ng 8.2 and kubuntu 22.04 without any modifications to the VM (just had to create xorg.conf so SDDM would use correct display)

now Wayland tries to use 1234:1111 (that's how virtual GPU is showing up in the VM when vga=on ) or "Cirrus" 1013:00b8 (when vga is disabled), I'm still trying to figure out how to tell Wayland to use nvidia

I'm using Nvidia A4000, no iGPU

acebmxer

I was able to get nvidia rtx 4070 passed through to a windows VM and Ubuntu server. Need to make sure you passing all the devices associated with the GPU, specifically the Audio controller on the GPU.

These were on AMD systems.

miltador

Good to see some successes here, no luck for me with Intel N100 iGPU though
Both Linux and Windows.

kubuntu-newbie

@acebmxer in my case it appears to be a Wayland issue, using the wrong card. I was hoping that I could disable virtual gpu all together

curious though that disabling virtual GPU just changes the device id in the vm from 1234:1111 to "cirrus" - that seems to be a bug in XCP-ng

acebmxer

@kubuntu-newbie

Yeah i didnt do anything with disabling the vgpu. Just a standard Win 11 or ubuntu VM and attached the nivdia gpu.

olivierlambert

IIRC, Intel iGPU is very different than a discrete GPU from a system perspective, that might explain your problem.

kubuntu-newbie

@olivierlambert
it is still not clear to me how changing vga: std to vga: cirrus in platform parameter of a vm constitute turning off vga (this is what happens when you turn off vga in Xen Orchestra)

manually setting vga to off in platform produces the same result as vga: cirrus, VM still sees "VGA compatible controller [0300]: Cirrus Logic GD 5446 [1013:00b8]"

is there a way to completely turn off creating this virtual VGA in the VM?

kubuntu-newbie

@olivierlambert

and removing key vga from platform all together has the same result as switching to cirrus, there is still a virtual vga Cirrus Logic in the VM

olivierlambert

Because it's the default you cannot bypass. @Team-XAPI-Network could confirm

kubuntu-newbie

@olivierlambert

thank you! I did a little digging, but I could be looking at the wrong place, according to:

https://github.com/xcp-ng/qemu-dp/blob/master/qemu-options.hx

there is an option vga: none

I tried setting this manually via "xe vm-param-set platform:vga=none " and I still get Cirrus PCI device created in the VM

Is there something else that needs to be done to have no virtual VGA in the VM at all?

psafont

@kubuntu-newbie I've looked at the code. The intel passthrough seems to not care about the value of the vga key in platform, and instead it expects the key igd_passthrough to be set to true.

https://github.com/xapi-project/xen-api/blob/9eb5f9f9f3c742c0c0b691098e1dafc02e40c856/ocaml/xapi/xapi_xenops.ml#L390