Hi guys,

Starting with a screenshot as I'm planning on creating a Rocky Linux VM machine for Davinci Resolve:

 

When Unraid put out beta 2 of 7.1, they explained that by enabling the Nouveau driver and then selecting "Virtual" in graphics Card section and setting the VM console driver field to "Virtio(3d)" you could virtualize the GPU so the VM machine could use it while the rest of unraid could use it. What I want to know is can I expand this to all the GPUs I have in my system? I have this on my testing machine so if a dev needs me to destroy (virtual chops and breaks) in the system to test this to get it working, do let me know. After setting up the initial virtual GPU, the other settings look like they are taking a GPU away from the rest of unraid which is what I don't want, I would want to virtualize all of them.

 

Regarding my end goal here, upfront is to simply be able to have the linux OS be able to fully use ALL my nvidia GPUs without forbidding their usage with the rest of my unraid system. I should be able to boot the linux system up, do my DaVinci Resolve work and rendering, close the application and then have a docker container use the GPUs for other work without haZving to detach or reattach stuff. I have achieved this with Ubuntu in a docker container.

 

I currently have on my other unraid machine, Ubuntu in a docker container and went through months of painful work to get the Ubuntu Container to work fully with DaVinci Resolve where it can use ALL the Nvidia GPUs, and fully function (encode h.264, h.265) with the timeline display working. It works a lot of the time but I sometimes hit limits that could be related to either the nature of the container (activation resets) or the power of my cards (I encode 8k HDR). I'm hoping that I can achieve this setup in a full fledge Linux VM running Rocky Linux which is DaVinci Resolve's recommended linux OS while still allowing my GPUs to be used by the rest of Unraid resources (Not needing to detach and re-attach).   

After some adjustments, I am only able to get the very first GPU working. If I add any more, the rocky linux OS is never able to boot.

 

Really hoping to see some progress on this with the actual NVIDIA drivers someday

10 hours ago, Velazquez said:

After some adjustments, I am only able to get the very first GPU working. If I add any more, the rocky linux OS is never able to boot.

 

Really hoping to see some progress on this with the actual NVIDIA drivers someday

It is not virtualization as in carving the card up. Virgl3d just routes OpenGL work streams to the GPU to process, similar to the way dockers would. Are 3060s individual cards?

 

The process creates the egl-headless for the render, I don't know if multiple can be added but will check. Please share diagnostics.

 

How would you use the multiple GPUs in the VM?

 

    <graphics type='vnc' port='-1' autoport='yes' websocket='-1' listen='0.0.0.0' sharePolicy='ignore'>
      <listen type='address' address='0.0.0.0'/>
    </graphics>
    <graphics type='egl-headless'>
      <gl rendernode='/dev/dri/by-path/pci-0000:00:02.0-render'/>
    </graphics>
    <audio id='1' type='none'/>
    <video>
      <model type='virtio' heads='1' primary='yes'>
        <acceleration accel3d='yes'/>
      </model>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x1e' function='0x0'/>
    </video>

 

 

Looks like only a single GPU can be supported.