Issues with NVIDIA Quadro M620 Mobile Passthrough on HP Z2 Mini – Need Advice

[theedude82]

Hi everyone,

I have a PVE cluster with three identical HP Z2 Mini computers (old CAD computers from my job i was gifted)—each with an NVIDIA Quadro M620 Mobile card. My goal is to pass through the GPU to an Ubuntu server 24.04 VM so I can run Plex for hardware-accelerated transcoding. Currently plex is running in a old i5 intel nuc.

What I’ve Tried So Far:

Custom ROM:
– I’m using the TechPowerUp ROM (230821.rom) placed in /usr/share/kvm/

Host PCI Settings:
– I’ve experimented with various hostpci0 configurations (e.g., using x-vga=on versus x-vga=off, and toggling rombar=0 vs. rombar=1), both with and without the ROM file.

CPU Flag Tweaks:
– I attempted to pass extra CPU options to hide virtualization (using -cpu host,kvm=off via the args field) to help convince the NVIDIA driver it’s on bare metal.

Observed Errors:
– Inside the VM, nvidia-smi always reports “No devices were found.”
– The guest’s kernel logs show errors like:
  NVRM: GPU 0000:01:00.0: Failed to copy vbios to system memory
  RmInitAdapter failed! (0x30:0xffff:1098)

I’ve tried every combination I could think of—but the mobile GPU still refuses to initialize properly. It seems that many mobile GPUs (including this Quadro M620 Mobile) don’t expose a usable UEFI VBIOS for VFIO passthrough.

My Hardware Details:

Hosts: HP Z2 Mini computers (three in a cluster)
GPU: NVIDIA Quadro M620 Mobile (one per host)
VM: Ubuntu server 24.04 (target for Plex transcoding)
My Question:
Has anyone successfully passed through a mobile GPU like the Quadro M620 Mobile on an HP Z2 Mini?

Has anyone managed to extract or use a valid UEFI-compatible VBIOS for this card (perhaps from a Windows install using GPU‑Z or a similar tool)?
Are there any additional configs or known limitations with mobile GPUs in Proxmox that I might be overlooking?
At this point, I’m considering moving to a system with a desktop-class GPU—but I’d love to learn if there’s a known workaround for these mobile GPUs.

Thanks in advance for any insights or suggestions. I’m happy to provide additional logs or configuration details if needed!

Erik

 

[szaboldi]

Hey, having the same problem. I spent almost all day with this. Did you manage to get it work?

 

[theedude82]

I could not get it working and dont think I'll be able to. Sucks

 

[szaboldi]

What do you think if i install windows on it extract the VBIOS and use that would it work or at this point it's just wasting time? Or would it be the same as the TechPowerUp ROM (230821.rom) ?

 

[theedude82]

When i asked chatgpt for help, it suggested i try that, but I don't have any Windows computers. I have a feeling it wont work either but if you have the resources i'd say try it out and see what happens.

 

[Ugh_noob]

I used to have it working as a was using it with a ollama ct. I went to try to update drivers remotely and something happened.

Causing me to lose access as well as other functionality currently trying to set it up again to use with a Windows Vm. Had to redo my cluster.

If I remember correctly, I had driver version 570. I had to install the drivers, Cuda and other supporting things [don’t quite remember] on the host then install the same driver on the vm/ct. Then running “nvidia-smi” I got GPU details both on node and vm level.

I remember creating documentation on the process to try to help but it was all over the place as it took hours to setup. Will post it when I have time.

Reference material: https://digitalspaceport.com/proxmox-lxc-gpu-passthru-setup-guide/

 

[zenowl77]

Since its quadro you may have better luck with vGPU, which would also let you pass up to 4x512mb profiles to multiple VMs (just not to LXCs)

polloloco vgpu-proxmox

Quadro shouldnt need licenses or unlock, theres places online you can find the KVM drivers without registering for a trial too.

(You probably want 16.9 / 535 or lower as its more work to get the newer drivers working on pascal / gtx 1000 or lower)

Edit: by the way, in general a desktop class GPU and newer GPU would be better in general for transcoding and support all around, even if it isnt the newest, I would probably go for a GPU with h.265/HEVC or AV1 support. Even the pascals have hevc support and they are a decade old now, thats what i use i have a Tesla P4, (and a gtx 1050 ti & gtx 1060 not in my server) the only newer card i have is a Arc A310 just for the AV1 support, they are cheap enough too for newer cards but they aren’t good for much else beyond transcoding.