Hey all, I have had a couple random issues starting to creep up on my unraid install and trying to figure out what is going on. First, I keep getting the Nvidia driver to go into an unresponsive state. nvidia-smi says no devices. GPU wont show in settings or in the dashboard widget. Reboot brings back the device for between 1 and 7 days. Within the last week Fix Common Problems kicked off a error stating that there was hardware issue (Ryzen 5800x). I took a look at the diagnostic logs but there wasn't anything that I could glean from the logs. Lastly, my rig locked and the DRAM light kicked on for the motherboard. Seems like the whole thing is going downhill.

I am running a Memtest currently but looking at the community to help me understand other things I can test. Seems like these issues are related but since we are taking ram, gpu, and cpu I am having a hard time focusing on what component could be the issue.

MCE hardware Error from logs

Jun 17 18:23:18 Oldtown kernel: mce: [Hardware Error]: Machine check events logged
Jun 17 18:23:18 Oldtown kernel: [Hardware Error]: Corrected error, no action required.
Jun 17 18:23:18 Oldtown kernel: [Hardware Error]: CPU:1 (19:21:0) MC20_STATUS[-|CE|-|AddrV|-|-|CECC|-|-|Scrub]: 0x850fc389c08500af
Jun 17 18:23:18 Oldtown kernel: [Hardware Error]: Error Addr: 0x0000000000000000
Jun 17 18:23:18 Oldtown kernel: [Hardware Error]: IPID: 0x0000000000000000
Jun 17 18:23:18 Oldtown kernel: [Hardware Error]: Bank 20 is reserved.
Jun 17 18:23:18 Oldtown kernel: [Hardware Error]: cache level: L3/GEN, tx: RESV

oldtown-diagnostics-20250626-0915(latest).zip oldtown-diagnostics-20250618-0759(hardware error).zip

Make sure this has been taken care of:

https://forums.unraid.net/topic/46802-faq-for-unraid-v6/page/2/#findComment-819173

Thanks for the reply @JorgeB .

Current BIOS Settings:

Aura - Off

DDR - 2666 for 2x16GB

SVM - Enabled

IOMMU - Enabled

HD Audio - Off

Onboard Lan - Off (I have a 10Gbit network card)

WiFi - Off

Bluetooth - Off

Fans - Turbo

I thought that CStates were largely fixed so I didn't do any CState / idle power management settings. I just went in and turned back on Power Supply Idle to be Typical Current Idle.

Would this cause issues with DRAM and the Nvidia GPU disappearing or am I troubleshooting multiple problems?

Edited June 26, 20251 yr by kbareis

It can, also see that part about slightly increasing voltage for 5000 CPUs, it can also help with stability issues in some cases

I read the links posted

https://wiki.archlinux.org/title/Ryzen#Random_reboots and https://community.amd.com/t5/pc-processors/ryzen-5900x-system-constantly-crashing-restarting-whea-logger-id/td-p/423321/page/84

Ended up also disabling PBO and CBP. Will run for a few days and circle back to the thread

Nvidia GPU dropped off again. Less than 24hrs uptime. Attaching screenshots and updated logs

oldtown-diagnostics-20250627-0815.zip

Edited June 27, 20251 yr by kbareis

So I think I just stumbled upon my issue. The VM I am running appears to be grabbing the Nvidia GPU when my docker images are backed up and upgraded over night. By rebooting the VM, the gpu comes back and is viewable via the Dashboard and nvidia-smi. However, the docker images that want to use it cannot and must be rebooted as well.

So the final question is, how do I force the VM / VM manager to not steal the GPU? The GPU for the VM is set to virtual already.

The VM should only use the GPU if it's on the XML.

Cross posted on the thread for the Nvidia driver. https://forums.unraid.net/topic/98978-plugin-nvidia-driver/page/198/#findComment-1564122

If you take a look at the screen recording, you can see the graphics driver for the VM is virtual but as soon as the VM is shut down, the GPU comes back and is available for the system again. This seems to happen overnight when Backup/Restore Appdata is backing up my containers.

oldtown-diagnostics-20250628-1220.zip

14 hours ago, Wedgetastic said:

the GPU comes back and is available for the system again

Try binding the GPU to vfio-pci, that should prevent it from ever being used by the driver/OS

I ended dropping back to the production driver and the GPU is no longer being grabbed by the VM