Hi UNRAIDers

 

Does UNRAID support the passthrough of Tesla cards (Tesla P4) for split workloads across multiple VMs? I assume the card will just appear as 1 GPU in the list of PCI devices? Would be really cool to be able to spread vGPU workloads rather than dedicate 1 GPU for every VM. Any ideas people? Thank you.

Hey mikeyosm,

 

I'm looking for the same. Recently bought two P4s and I would like to use it for 4 VMs. I saw a great tutorial from craftcomputing to do this on Proxmox with M40s but I'm not knowledged enough to translate the tutorial to unraid (if even possible).

Link to the tutorial: 

 

Please let me know if you found somthing on this topic!

Same here

On 11/12/2022 at 8:04 AM, csard said:

Hey mikeyosm,

 

I'm looking for the same. Recently bought two P4s and I would like to use it for 4 VMs. I saw a great tutorial from craftcomputing to do this on Proxmox with M40s but I'm not knowledged enough to translate the tutorial to unraid (if even possible).

Link to the tutorial: 

 

Please let me know if you found somthing on this topic!

 

I used this with success on my P4's

 

https://www.lxg2016.com/55875.html

Can you share a bit more info on version used or other thing seem to be getting error

 

I have a Tesla P40 and i actually bring it to work!

 

So you are able to split your GPU and bring it to the VM. 

And i use this one https://github.com/stl88083365/unraid-nvidia-vgpu-driver 

 

Its takes a little bit to understand how it works but after then it was really easy to use!

any info on these would be appreciated as if you want 4 or 8 instance of windows and then how to you pass it to the vm

 

## Modify the following variables to suit your environment

WIN="2b6976dd-8620-49de-8d8d-ae9ba47a50db"  #do you need one per vm or just one

UBU="5fd6286d-06ac-4406-8b06-f26511c260d3"

MDEVLIST="nvidia-65" #say you want mode 64 that is 4 instance with 2gig of ram on teslaP4 with 8gig

 

 

also how to configure the vm onces de mdev are created

 

edit: From waht i am readin you need to manually edit the vm in xml vieu after installing windows probably best.  THe UUID is the UUID of the mdev you have created with the script.  YOu need to create one mdev/UUI per VM and need to use the same mdev mode for all.

 

SO in my case I have created moded the script to make 4 UUID that use in my VMs.  I am using a tesla P4 and mode 64.  But not using the unlock and not using the profile overide as the T4 is supported directly.

 

<hostdev mode='subsystem' type='mdev' managed='yes' model='vfio-pci' display='off'>
      <source>
        <address uuid='2b6976dd-8620-49de-8d8d-ae9ba47a50d1'/>
      </source>
      <alias name='hostdev0'/>
      <address type='pci' domain='0x0000' bus='0x04' slot='0x00' function='0x0'/>
    </hostdev>

Edited March 7, 20242 yr by AppleJon

For other finding this thread these link help me figure things out more

 

proxmox but section on vGPU and UUID and MDEV creation is helpfull

https://wvthoog.nl/proxmox-7-vgpu/

 

this for the bottom section on mediated device and getting them assigned to the vm

https://ubuntu.com/server/docs/gpu-virtualization-with-qemu-kvm

Edited March 8, 20242 yr by AppleJon