Hardware: Z790 + RTX 5060 Ti. This issue occurs on Unraid 7.4-beta3, 7.3.3-rc2, and also versions 7.3.2 and older. When passing through the GPU audio device, WebUI hangs on all tested Unraid versions.

While the WebUI is hung, I can create new SSH connections via Xshell. Commands like reboot, poweroff, and powerdown have no effect on Unraid. After running poweroff, the active Xshell SSH connection drops, and new SSH connections cannot be created. My previously opened ip/webterminal/ttyd/ and ttyd sessions remain alive without disconnecting. Running df -h shows all disks are still mounted, but any VM-related commands hang when executed.

link to this discussion thread

win11 xml.txt syslog syslog-previous lynxserver-diagnostics-20260929-0057.zip

Here is a clue for reference. My friend uses Z890 + RTX 5060 Ti with GPU audio passed through to the VM. Unraid works normally after VM shutdown. It stays stable on Unraid 7.3.2 and older versions. On his system, the GPU and its audio device are in separate IOMMU groups.

Are both the GPU and its audio part bound to vfio?

Thanks, the previous syslog captures the shutdown failure:

BUG: kernel NULL pointer dereference, address: 000000000000041c

RIP: 0010:nv_audio_dynamic_power+0x8e/0x100 [nvidia]

Both the GPU and its audio device report successful resets. NVIDIA then reclaims the GPU for the host and crashes while restoring the display and audio state. The log also records qemu-event exiting with interrupts disabled. This explains why VM management can stop responding while some terminal sessions remain available.

The signature matches NVIDIA issue 1141 (https://github.com/NVIDIA/open-gpu-kernel-modules/issues/1141), which remains open without a linked fix.

This is separate from the startup problem in your other thread.

Since you need the GPU for Docker, return to your virtual-audio configuration without passing through the NVIDIA audio device for now. Please confirm whether that configuration also allows normal VM shutdown and subsequent Docker GPU use.

Binding both NVIDIA functions to VFIO at boot would avoid returning the card to the NVIDIA host driver. However, Docker could no longer use it. We do not yet have a verified fix that preserves physical audio passthrough and reliable GPU return to Docker.

2 minutes ago, Kilrah said:

Are both the GPU and its audio part bound to vfio?

No. I need to make the NVIDIA GPU available for both Unraid and Docker.

5 minutes ago, JorgeB said:

Thanks, the previous syslog captures the shutdown failure:

BUG: kernel NULL pointer dereference, address: 000000000000041c

RIP: 0010:nv_audio_dynamic_power+0x8e/0x100 [nvidia]

Both the GPU and its audio device report successful resets. NVIDIA then reclaims the GPU for the host and crashes while restoring the display and audio state. The log also records qemu-event exiting with interrupts disabled. This explains why VM management can stop responding while some terminal sessions remain available.

The signature matches NVIDIA issue 1141 (https://github.com/NVIDIA/open-gpu-kernel-modules/issues/1141), which remains open without a linked fix.

This is separate from the startup problem in your other thread.

Since you need the GPU for Docker, return to your virtual-audio configuration without passing through the NVIDIA audio device for now. Please confirm whether that configuration also allows normal VM shutdown and subsequent Docker GPU use.

Binding both NVIDIA functions to VFIO at boot would avoid returning the card to the NVIDIA host driver. However, Docker could no longer use it. We do not yet have a verified fix that preserves physical audio passthrough and reliable GPU return to Docker.

Okay, thank you very much. I'll stick with the virtual audio configuration to keep the GPU accessible for Unraid and Docker.