NVIDIA GPU Passthrough - working then: Failed to initialize NVML: Unknown Error

stormy27

Active Member
Feb 28, 2019
14
1
43
Hi all!

My problem: GPU Passthrough (GTX1070) is working some minutes then proxmox disable/deactivate it! Only reboot can re-enable for short amount of time....


Using PROXMOX 8 server I'm trying to enable my NVIDIA GTX 1070 in passthrough mode to one of my VM.
(followed the thread here: https://gitlab.com/polloloco/vgpu-proxmox )

Everything was correctly done, with confirmed 'nvidia-smi vgpu' enabled and so on under PROXMOX. I used the patch driver also, latest release 16.

2023-10-23 16_07_33-pve - Proxmox Virtual Environment — Mozilla Firefox.png

2023-10-23 17_33_54-__DATASTORAGE_Akaoma_Division IT_HP GEN8 380p_VM Passthrough GPU problem_2...png


It's working on the VM Windows is correctly detecting the NVIDIA card
BUT after some short time the driver TOTALLY disappears directly on PROXMOX.


2023-10-23 16_16_11-pve - Proxmox Virtual Environment — Mozilla Firefox.png
root@pve:~# nvidia-smi vgpu Failed to initialize NVML: Unknown Error

root@pve:~# mdevctl types gives no answer


And under the (previous) working VM with GPU passthrough the NVIDIA is no more showing MEDIATED "YES" .....
2023-10-23 17_34_56-Greenshot image editor.png



Of course everything was fine before....

My motherboard is using a RYZEN 5700G CPU, 128GB RAM. IOMMU enabled and all BIOS options (SVM etc)

Impossible to retrieve a working NVIDIA system. I've read many many things but I feel lost.
Only rebooting the PROXMOX server and I find again the working answer to the commands "nvidia-smi", "mdevctl types" and "nvidia-smi vgpu".

For a short amount of time again.... because it's staring again.

I don't know what to do know. So weird!


Could you please help me?? How can I test ? Check ? Thanks ! :)


I found this in the /var/log/kernel file
2023-10-23 16_17_13-pve - Proxmox Virtual Environment — Mozilla Firefox.png



Here are some screencopy showing the GPU NVIDIA correctly shown under my Windows VM using the GPU passthrough (here is an AI object recognition project)
2023-10-23 17_16_00-QEMU (BlueIris) - noVNC — Mozilla Firefox.png
 

Attachments

  • 2023-10-23 16_09_19-pve - Proxmox Virtual Environment — Mozilla Firefox.png
    2023-10-23 16_09_19-pve - Proxmox Virtual Environment — Mozilla Firefox.png
    44.5 KB · Views: 6
Last edited:
  • Like
Reactions: jreh