Proxmox shared pass through Quadro to multiple VM

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

ReturnedSword

Active Member
Jun 15, 2018
526
235
43
Santa Monica, CA
I got into Proxmox recently with my TMM. I’m able to pass through a Quadro P1000 fine to a Debian VM, though it would be nice to share the GPU with more than one VM.

I recall this can be done with Quadro, and GeForce (with hacked configuration) using nVidia vGPU. I’m having the darnest time figuring out how to get this done, however. There just isn’t much info out there that I can find regarding this.

Is this even possible in Proxmox? Or only RHLE?

If this isn’t possible in Proxmox, does anyone have some experience to share about sharing a GPU in VMWare?
 
  • Like
Reactions: MrCalvin

damienr

New Member
Nov 20, 2015
23
5
3
New Zealand
vGPU_Unlock perhaps? GitHub - DualCoder/vgpu_unlock: Unlock vGPU functionality for consumer grade GPUs.

I had it running up to Kernel 5.12 I think, but beyond that it broke (and I didn't care enough to fix it).
Up until that stage, it was working in Proxmox 7 for me. 470.63 NVidia driver.

I used a combination of YouTube and other links to achieve it, on an AMD X570/3700X setup.
eg https://wvthoog.nl/proxmox-7-vgpu/

Hope that was what you were after. Good luck.

EDIT: Also a Quadro P1000. Used 2x 2G profiles and most prominent guest OS was Win10.
 

ReturnedSword

Active Member
Jun 15, 2018
526
235
43
Santa Monica, CA
Thanks @damienr for your reply.

It looks like there aren’t any working solutions for the current kernel. I’ll have to roll back my kernel version and try that out.

In your experience, did it seem pretty hacky? AFAIK in RHEL and ESXi nVidia vGPU is a lot more straightforward (even with the vGPU mod for unsupported GPUs).
 

ectoplasmosis

Active Member
Jul 28, 2021
117
53
28
How are you planning to present the vGPU-accelerated VMs to the end user?

Is there a way to enable the physical display outputs on a GPU and assign one to a VM?

If not, which method is the best to present the GUI to a user?
 

damienr

New Member
Nov 20, 2015
23
5
3
New Zealand
Thanks @damienr for your reply.

It looks like there aren’t any working solutions for the current kernel. I’ll have to roll back my kernel version and try that out.

In your experience, did it seem pretty hacky? AFAIK in RHEL and ESXi nVidia vGPU is a lot more straightforward (even with the vGPU mod for unsupported GPUs).
That's a good question. Kind of, kind of not. Looking back, no more 'hacky' than the process to make GPU passthrough work.
If memory serves, a few options I had to perform then, now come standard on Proxmox. Like GRUB.
Guides are also more refined now, so steps are easier to take (just looking at that wvthoog URL again)

Can't really speak for the other OS's you mention as no personal experience.
For my use case, I found that I was only really using one guest with passthrough at a time - so regular passthrough was sufficient. No faffing with both host and guest drivers that way
 

ReturnedSword

Active Member
Jun 15, 2018
526
235
43
Santa Monica, CA
How are you planning to present the vGPU-accelerated VMs to the end user?

Is there a way to enable the physical display outputs on a GPU and assign one to a VM?

If not, which method is the best to present the GUI to a user?
The initial use case would be for a Debian based VM running Plex in Docker for transcoding.

Honestly, I haven’t thought of subsequent use cases yet. Possibly a graphics accelerated Windows VM. My intention for setting this up ahead of time is so I won’t have to fiddle with it later and disrupt any running services. In any case, any use would be headless. Any service I’ll be using will have a web interface. If I need to access the VM directly I’ll be doing that over ssh.