AMD's shader compiler on Windows on RDNA3+ is known to be extremely slow.
Doesn't really surprise me... I've had nothing but subpar experiences with AMD on Windows so far, except with VRR support, weirdly enough that aspect works better on AMD. For all the rest though, Nvidia on Windows is just a generally superior experience.
It's so weird though, seeing them struggle with Vulkan shader compiling, considering they kickstarted the whole thing initially with Mantle. Oh well, lesson learned I guess.
Because this bug is not fixed - https://github.com/doitsujin/dxvk/issues/4840#issuecomment-3994354911
Info from the first source: https://forums.guru3d.com/threads/dx9-dxvks-shader-compilation-seems-broken-in-rdna-4-gpus.459451/post-6402294
Because this bug is not fixed - [#4840 (comment)](https://github.com/doitsujin/dxvk/issues/4840#issuecomment-3994354911)
Info from the first source: https://forums.guru3d.com/threads/dx9-dxvks-shader-compilation-seems-broken-in-rdna-4-gpus.459451/post-6402294
This is super interesting thank you, I'll be waiting for that DXVK-GPLALL 3.x branch of yours then :)
If I'm not too crazy, I read that there's supposed to be a completely new and efficient non blocking real time shader compilation process since DXVK 3.0,
No. The new compiler fixed a couple of bugs that affected very few games and produces significantly better code but it won't magically speed up shader compilation in the driver. I'd where you get the idea that it's 'non blocking' from either.
From a users POV the new shader compilers shouldn't really make a difference.
DXVK 2.0 already had support for EXT_graphics_pipeline_libraries which allowed DXVK to compile slower temporary shaders when the game loads the original D3D shaders and then compile the faster shaders in the background as soon as possible.
That pretty much solved shader stutter in the vast amount of games.
If that doesn't work on AMD, the shader compiler in their Windows driver is simply too slow.
In that case there's just not enough time between the game 'sending us the shader code' and trying to draw triangles with it...
If I'm not too crazy, I read that there's supposed to be a completely new and efficient non blocking real time shader compilation process since DXVK 3.0, but while I can totally see that on Nvidia (several cards), on my AMD (RX 9070 XT) GPU, it feels almost like pre 2.0 days on Nvidia, this is on Black Mesa with the D3D9 fallback on Windows 10, using DXVK 3.0.2 (same on 3.0.1). I was wondering why is there such a big difference between the 2 vendors, is it just driver related?