I am facing a bizarre issue while trying to pass through my two Vega FE to a KVM-hosted windows 10 VM.
I am able to pass through the GPU, install the Radeon Pro drivers and start using it (either with synthetic benchmarks like Engine Valley or with Folding @ Home).
Randomly, however, the GPUs power state LED blink orange from the default blue, the fan goes suddenly to 100% and the VM locks completely, requiring a full host (yes, the whole server) shutdown and reboot to fix the weird issue.
Today I ran 1:30h effortlessly, and then I had to restart the system 3 time in less than an hour.
To avoid redundant information, here is my detailed setup was posted on the Proxmox forum: Vega Frontier passthrough help | Proxmox Support Forum
I am now running Nvidia GPUs in the exact same slots and nothing strange happens, allowing me to get the VM working with great stability.
For good measure I have disabled the ULPS state from the windows registry, with no success.
Can someone point me to the right direction to make my whole system work as intended?
Don't be afraid to get technical, since I am a system admin...