I've experienced a few issues in the past few days after binding my iGPU to VFIO and wanted to see if anyone has any insight on why this happened. The tl;dr: I bound my iGPU to VFIO, which caused RAM usage to unboundedly increase, a series of unexpected low disk space emails, and the web GUI became inaccessible.

 

I'm running Unraid 7.0.0 and have a Ryzen 7 5700G CPU, where I was hoping to make use of the integrated graphics in a Windows virtual machine. Here's what my IOMMU grouping looks like, and convinced myself that this was safe enough to do and wouldn't affect, for example, array hard disks. 

 

On Friday 14th March, I bound this group to VFIO and did a reboot on the system. I didn't realise at the time, but after this point the memory usage started to increase (see the output from the Glances Docker container that is fed into a separate Home Assistant instance below). By coincidence, I ended up doing another reboot on 17th which, which is where you can see the sharp drop in RAM usage. There's 64GB of memory in my server so it's actually using quite a lot -- or at least much more than my system usually does. I never did manage to get the GPU pass-through to work, so there was no additional RAM usage on that new VM.

 

 

Then, on Wednesday 19th, three weird things happened (at the peak of RAM usage in graph above):

• Had a flurry of emails over the course of 10-or-so minutes on a relatively newly-installed disk to say that disk6 was 100% full, despite it being more like 50%  
• WebGUI went down, giving a 500 error on browsers where I was already logged in, and the normal login page with my server name missing on browsers -- the login button on that page would do nothing
• All Docker containers continued to work as normal (arguably not weird, but still a little bit weird, right?!) 

 

I panicked a little and pressed the power button on my server to initiate a shutdown, but in hindsight I should've tried to recover the logs via SSH first. Does anyone have any ideas on what may have happened here, and if there's something I have done wrong here? Thanks in advance for any help! 🙂

while possible binding the IGPU is not recommend as this esentail pulls apart the CPU. And prevents and removes some needed CPU access to the chip.

 

Binding the integrated GPU (iGPU) of your Ryzen 7 5700G to VFIO for passthrough in Unraid can lead to system instability and resource management issues. Users have reported challenges such as increased memory usage, storage anomalies, and web GUI inaccessibility after attempting iGPU passthrough.

 

? AMD Reset bug...

 

Potential Causes:

AMD iGPU Reset Bug: AMD's iGPUs are known to have reset bugs, making them problematic for passthrough scenarios. This bug can prevent the iGPU from properly resetting between VM reboots, leading to system instability.

 

Resource Allocation Conflicts: The iGPU shares resources with the CPU, and binding it to VFIO might interfere with system processes that rely on the iGPU, resulting in unexpected behavior such as increased RAM usage or disk space issues.

 

Recommendations:

Avoid iGPU Passthrough: Given the challenges and potential system instability, it's advisable to avoid passing through the iGPU. Instead, consider using a discrete GPU for VM passthrough needs.

 

Monitor System Resources: Regularly monitor your system's RAM and disk usage to identify and address any anomalies promptly.

Community Consultation: Engage with the Unraid community forums to share experiences and seek advice. Others may have encountered similar issues and can offer insights or solutions.

 

By refraining from binding the iGPU to VFIO and opting for a dedicated GPU for passthrough, you can maintain system stability and avoid the complications you've experienced.

 

Can it be done yes, but you invite trouble... Not all igpu work corectly...

 

 

 

I would advise to not vfio your IGPU for the sake of the CPU...

Edited March 22, 20251 yr by bmartino1
typo - Data

Thanks @bmartino1 for that advice, much appreciated! I think I'm going to abandon hope of making use of the iGPU.

 

Going back to the issue I was seeing, I've figured out that the issue was to do with the Mover Tuning plugin writing out large .list files, which filled up /tmp. File logging is disabled so I'm not sure why this was happening, nor am I sure why I was getting disk full warning emails, but at least I know what was causing the large RAM usage now.