Mesa 23.3.1 Rolls Out Fixes From D3D12 To RADV

[phoronix]

Phoronix: Mesa 23.3.1 Rolls Out Fixes From D3D12 To RADV

It's been two weeks since Mesa 23.3 was punctually released and this week has been succeeded by the Mesa 23.3.1 point release with the first batch of bug/regression fixes...

https://www.phoronix.com/news/Mesa-23.3.1-Released

[Brisse]

Anyone else having problems with RadeonSI and American Truck Simulator or Euro Truck Simulator 2? I've yet to try 23.3 (will do later today) but this has been bugging me for a while, probably all 23.* releases, and perhaps even further back. Game crashes within a few minutes and takes the entire desktop down with it. Computer locks up mostly, except I can ssh into it and kill the process from there or reboot.

R9 Fury, aka. Fiji, aka. gfx803.

Zink works fine. Nvidia works fine (yes I have an RTX2070S running on the proprietary driver in the same PC).

Problem is exclusively happening with RadeonSI.

Edit: Still crashing after updating to today's mesa release. Also tried disabling mesa gl-threading but still crashing.
Maybe the same issue as this: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9593

[Danny3]

Anyone else having problems with RadeonSI and American Truck Simulator or Euro Truck Simulator 2? I've yet to try 23.3 (will do later today) but this has been bugging me for a while, probably all 23.* releases, and perhaps even further back. Game crashes within a few minutes and takes the entire desktop down with it. Computer locks up mostly, except I can ssh into it and kill the process from there or reboot.

R9 Fury, aka. Fiji, aka. gfx803.

Zink works fine. Nvidia works fine (yes I have an RTX2070S running on the proprietary driver in the same PC).

Problem is exclusively happening with RadeonSI.

No with Euro Truck Simulator 2, that I'm actually playing these days.
But I run the Windows version so I guess it's running through Proton on RADV with Mesa 23.2.1 on my AMD RX 560 (GFX8).
I rather not test the native version if if the bug is so bad, sorry.

[Kjell]

Is it just me or are plenty of games stuttering with MESA? Some are buttersmooth while others aren't. The issue isn't tied to DX11/DX12 and neither power managment nor WM/DE

One very particular example is: High On Life

[andre30correia]

i update to this new version and the things are better after faulty 23.2 realease, now the games im playing are again working well, even firefox becomes more fast in javascript renders using webgpu and hw. canonical should open a exception and upgrade their mesa in mantic to 23.3.1 because the bad 23.2.1 will be backported to the last lts and its really bad realease with no point releases to become stable a complety disaster i hope the maanger of that realease no more take control of another one

[QwertyChouskie]

Is it just me or are plenty of games stuttering with MESA? Some are buttersmooth while others aren't. The issue isn't tied to DX11/DX12 and neither power managment nor WM/DE

One very particular example is: High On Life

https://gitlab.freedesktop.org/drm/amd/-/issues/2516#note_1874760

[aufkrawall]

Is it just me or are plenty of games stuttering with MESA? Some are buttersmooth while others aren't. The issue isn't tied to DX11/DX12 and neither power managment nor WM/DE

There is a nasty performance regression in amdgpu kernel driver since 6.2 (?) that apparently causes absurd overhead with radv_zero_vram=true (.drirc default for DXVK/VKD3D-Proton).
As a workaround, radv_zero_vram=false (can also conveniently be mis-used as an env var) often reduces stutter to 0, but can theoretically cause visual corruption at any time (though in practice, it often doesn't at all).

Edit: Not sure if it actually is a regression of 6.2, some issues have been mixed up.

[Kjell]

https://gitlab.freedesktop.org/drm/a...6#note_1874760

Thanks for bringing it up!
It makes stuttering less intensive

(I'm running mesa-git and git of dxvk+vkd3d)

[Kjell]

There is a nasty performance regression in amdgpu kernel driver since 6.2 (?) that apparently causes absurd overhead with radv_zero_vram=true (.drirc default for DXVK/VKD3D-Proton).
As a workaround, radv_zero_vram=false (can also conveniently be mis-used as an env var) often reduces stutter to 0, but can theoretically cause visual corruption at any time (though in practice, it often doesn't at all).

Edit: Not sure if it actually is a regression of 6.2, some issues have been mixed up.

The core issue doesn't seem to be related to memory managment as I've tried the kernel patch 0001-drm-amdgpu-WIP-always-wipe-on-release.patch with no result (the same patch solved vram zeroing in Cyberpunk for everyone else) :/

I'll try this ENV in a second and post a update

[Kjell]

As a workaround, radv_zero_vram=false (can also conveniently be mis-used as an env var)

Just tried radv_zero_vram=false as a environment variable when starting the game but it made no difference either :/