Proxmox 8.4 qemu patch for vGPU

I’m currently pulling my hair out trying to get an A40 virtualized on Proxmox 8.4, kernel 6.8. I can see the virtual vGPUs. I had no issue getting vGPU working for Proxmox 9.0.6, I simply enabled bios features, blocked the nouvea driver, rebooted and could see the mediated devices within the resource mapping in Proxmox, however the drivers that are compatible with Proxmox 9.0.6 / Debian 13 utilise CUDA 13, and the software I am running within the VM only supports CUDA 12.2. I tried to install an older driver but unfortunately they wouldn’t build with the kernel modules included in the Proxmox 9.0.6 build, and no way to downgrade.

So I thought i’d go back to an older version of Proxmox, supported kernel, older Nvidia driver, simple (I THOUGHT!)

It appears a lot has changed in the way the host presents the mediated devices and with 8.4 I just cannot get this working. My research says that I need a patch for qemu from nvidia, does anybody know where I can download this? I have looked in my NLP dashboard but none of the driver package include a patch for qemu.

Current install Driver version: NVIDIA-Linux-x86_64-570.172.07-vgpu-kvm.run

Snippet of the nvidia-smi -q output:

root@mep-proxmox01:~# nvidia-smi -q

==============NVSMI LOG==============

Timestamp : Thu Oct 2 15:21:35 2025
Driver Version : 570.172.07
CUDA Version : Not Found
vGPU Driver Capability
Heterogenous Multi-vGPU : Supported

Attached GPUs : 1
GPU 00000000:17:00.0
Product Name : NVIDIA A40
Product Brand : NVIDIA
Product Architecture : Ampere
Display Mode : Enabled
Display Active : Disabled
Persistence Mode : Enabled
Addressing Mode : N/A
vGPU Device Capability
Fractional Multi-vGPU : Supported
Heterogeneous Time-Slice Profiles : Supported
Heterogeneous Time-Slice Sizes : Supported
Homogeneous Placements : Not Supported
MIG Mode
Current : N/A
Pending : N/A
Accounting Mode : Enabled
Accounting Mode Buffer Size : 4000
Driver Model
Current : N/A
Pending : N/A
Serial Number : 1321723046272
GPU UUID : GPU-3da99dc1-6f59-ded9-fc4c-01531004705c
Minor Number : 0
VBIOS Version : 94.02.5C.00.0F
MultiGPU Board : No
Board ID : 0x1700
Board Part Number : 900-2G133-0100-130
GPU Part Number : 2235-895-A1

Proof that the host can see the vGPU profiles:

cat /sys/bus/pci/devices/0000:17:00.4/nvidia/creatable_vgpu_types
ID : vGPU Name
555 : NVIDIA A40-1B
556 : NVIDIA A40-2B
557 : NVIDIA A40-1Q
558 : NVIDIA A40-2Q
559 : NVIDIA A40-3Q
560 : NVIDIA A40-4Q
561 : NVIDIA A40-6Q
562 : NVIDIA A40-8Q
563 : NVIDIA A40-12Q
564 : NVIDIA A40-16Q
565 : NVIDIA A40-24Q
566 : NVIDIA A40-48Q
567 : NVIDIA A40-1A
568 : NVIDIA A40-2A
569 : NVIDIA A40-3A
570 : NVIDIA A40-4A
571 : NVIDIA A40-6A
572 : NVIDIA A40-8A
573 : NVIDIA A40-12A
574 : NVIDIA A40-16A
575 : NVIDIA A40-24A
576 : NVIDIA A40-48A

Can anybody point me to the patch for qemu or any other information that could help?

Hi,

from the description above I cannot see what is not working right now. You even showed a screeshot where all the MDEV devices are shown properly.
In addition, I’m not aware of any patch necessary.

Can you create a support ticket with NVES if you have valid licenses?