GPU Passthrough to VM Question

[panzerscope]

Hey guys,

I have a single GPU in my TrueNas Scale server (TrueNAS-SCALE-22.12.2). How do I set it to pass through to a VM ? I have 3 running VM's which I would like to take advantage of the GPU. I tried editing the VM to add in the GPU but I get an error "At least 1 GPU is required by the host for its functions, with your selection there is no GPU for the host to function"

I have read a couple of articles, but want to confirm that you need two GPU's if you wish to passthrough to VM ? My current GPU as stated is used by the overall system including my Plex App and I am guessing that is ultimately the issue.

Any advice is very much appreciated.

Regards,
P

 

[HoneyBadger]

Correct - in order to do full isolation of a GPU and pass it through to a VM as a PCIe device, the host OS (TrueNAS SCALE) still needs to have a GPU for itself, even if you aren't using it for containerized Apps such as Plex/Jellyfin.

I have 3 running VM's which I would like to take advantage of the GPU.

Note that only one VM will be able to be powered on and use the GPU at a time - there's presently no driver support for things like GPU-IOV or NVIDIA vGPU in SCALE (although I really ought to give it a shot on a test machine that I'm willing to blow up)

 

[panzerscope]

Correct - in order to do full isolation of a GPU and pass it through to a VM as a PCIe device, the host OS (TrueNAS SCALE) still needs to have a GPU for itself, even if you aren't using it for containerized Apps such as Plex/Jellyfin.


Note that only one VM will be able to be powered on and use the GPU at a time - there's presently no driver support for things like GPU-IOV or NVIDIA vGPU in SCALE (although I really ought to give it a shot on a test machine that I'm willing to blow up)

Thanks for the reply! I thought this would be the case that I would a secondary GPU. That being said, I think Scale would be that much better if you could have multiple VM's using a single GPU. We know that it is out their, just requires some testing and implementation, here is to hoping that happens one day.