gpu problem

[a1360550545]

pve version 8.1.3 Virtual machine: built cloudinit ubuntu20 When I use gpu graphics card pass-through, everything is normal, but after a while, I will find that when one of my graphics cards is mounted on the virtual machine, the virtual machine cannot start, but it can be replaced by another graphics card. It is broken. The graphics card cannot be used by any virtual machine until I restart the pve host. I am using mapping, but after the problem occurs, I cannot solve it even if I use the original device. I need help.

generating cloud-init ISO
kvm: ../hw/pci/pci.c:1637: pci_irq_handler: Assertion `0 <= irq_num && irq_num < PCI_NUM_PINS' failed.
TASK ERROR: start failed: QEMU exited with code 1

 

[a1360550545]

The log information queried is as follows:

vfio-pci 0000:85:00.0: Unable to change power state from D3cold to D0, device inaccessible

 

[leesteken]

Sounds like the GPU does not reset properly and can only be used once (per Proxmox host boot). It is not uncommon. Maybe there are work-around for your GPU, but you will have to reveal the make and model.

 

[a1360550545]

This problem occurred after using it for a period of time. My gpu model is rtx3090, and there are two of them. At first, both of them can be used normally, but after a while, only one can be used normally.

Sounds like the GPU does not reset properly and can only be used once (per Proxmox host boot). It is not uncommon. Maybe there are work-around for your GPU, but you will have to reveal the make and model.