Intel Arc VM pass through?

[10wuebc]

Does anyone know if the new Intel Arc GPUs have VM passthrough. I have a friend who has a plex server on a VM and he can't seem to get the Nvidia drivers to work with the pass through and was wondering if Intel ARC had that pass through ability to make it easier for him.

[Eigenvektor]

Passthrough isn't really a GPU capability. That's something the host does to simulate the card being attached to the VM.

 

The exception would be server cards that explicitly support virtualization and allow a single GPU to be shared between multiple machines without passthrough.

 

Some Intel iGPUs support virtualization: https://wiki.archlinux.org/title/Intel_GVT-g

 

If that's not the case for Arc (probable) then its down to the hypervisor to do passthrough correctly, same as with Nvidia

[10wuebc]

46 minutes ago, Eigenvektor said:

Passthrough isn't really a GPU capability. That's something the host does to simulate the card being attached to the VM.

 

The exception would be server cards that explicitly support virtualization and allow a single GPU to be shared between multiple machines without passthrough.

 

Some Intel iGPUs support virtualization: https://wiki.archlinux.org/title/Intel_GVT-g

 

If that's not the case for Arc (probable) then its down to the hypervisor to do passthrough correctly, same as with Nvidia

I guess what i meant to say is will intel / proxmox support passthrough day 1 with arc, without having to go through a big hassle.

[Eigenvektor]

3 hours ago, 10wuebc said:

I guess what i meant to say is will intel / proxmox support passthrough day 1 with arc, without having to go through a big hassle.

Gotcha. From my (limited) understanding the GPU manufacturer shouldn't really have to do anything to support it. The hypervisor hands the PCIe device into the virtual machine, making it appear like a physically attached card, which means it should work exactly the same.

 

From what I could find, Nvidia previously prevented their driver for consumer hardware from working in a VM (https://clayfreeman.github.io/gpu-passthrough/). This seems to have changed with newer driver versions (https://mathiashueber.com/passthrough-windows-11-vm-ubuntu-22-04/). So it's less a case of Nvidia did something to support it, but rather they stopped preventing it.