Ryzen 9 9950X3D iGPU (RDNA 2 Raphael)

[felixhauptmann]

Disclaimer: Im working with virt-manager/KVM instead of Proxmox but there should not be a difference.

• I have extracted the vbios and modified the GOPDriver from my MSI PRO X870-P WIFI Motherboard
• I can successfully boot the VM (OVMF) and do not get Error 43 when providing both romfiles
• RadeonResetBugFix "works" but I get weird visual glitches after the second boot (Windows 11)
• I also get spurious Devices changed (cpu or memory) or cmos have been cleared messages from my host BIOS

Related PR: #136

[felixhauptmann]

Just to document my own mistakes:
I did some cleaning up (found old udev rules for vendor-reset) and played around with BIOS settings (Resizable BAR/UMA size).

I have somewhat isolated the issue:

The flickering is always accompanied by a ton of AMD-VI IO_PAGE_FAULT messages.

This is my current BIOS setup

• Re-Size BAR Support disabled
• Above 4G Decoding enabled

First boot always works, then one of three things happen:

• It still works??
• Very Bad Visual glitches: dmesg shows IO_PAGE_FAULTs
• iGPU not working at all: No Page Faults

I feel like forcing 8 GB produces the least amount of issues/most reboots without things breaking but this is obviously very funky.

=> It seems like the iGPU has issues respecting the allocated shared VRAM

But isn't this the exact thing IOMMU is supposed to solve?

I Kinda feel like it might be firmware bug with my hardware. Or it could just be a stupid mistake on my side.

It would be awesome if someone with similar hardware could share their results.

[fcaronte]

What your proxmox version? I have same issue on my Ryzen 5 5625U on proxmox 9 with 6.14 kernel first boot work normally then from secont reboot i have screen glitch (using remotely gpu is working correclty)

For me actually I partial fix installing the old 6.8 kernel were seem to work better for me (some time stuck too). i have enabled 4g decode and bar but if for example i check on gpuz say rom bar disabled so maybe is not supported. i force 4gb of vram but i try with 16 with same result.

on new kernel i try to edit vendor reset to try support my cpu but not work same glith as with standard vfio

[felixhauptmann]

@fcaronte Please read my comments again, as I said, I am not using Proxmox but I'm pretty sure what were experiencing is not related to Proxmox.

[fcaronte]

Ok so you have a Linux host and use virt-manager to run a window VM with GPU, can you check host kernel version and try with an old 6.8 based kernel? As today on my server I stuck on 6.8 to keep working correctly the passthrough, sometimes when I see a new kernel update I try if maybe fix but actually the glitch persist, so I think for you will work same way if use an old kernel version.
About bios settings I try the ROM size but I don't see to much different and as for you have same log when is glitched.

@fcaronte Please read my comments again, as I said, I am not using Proxmox but I'm pretty sure what were experiencing is not related to Proxmox.