Mesa's RADV Radeon Vulkan Driver Continues Outperforming AMDVLK For RDNA2 Linux Gaming

[phoronix]

Phoronix: Mesa's RADV Radeon Vulkan Driver Continues Outperforming AMDVLK For RDNA2 Linux Gaming

With last week seeing AMDVLK 2022.Q2.3 released as an update to AMD's official open-source Vulkan Linux driver and it noting performance improvements, it was time for some fresh benchmarks of that driver up against Mesa's alternative "RADV" Vulkan driver. Here are some fresh benchmarks with an AMD RDNA2 GPU for seeing how RADV continues competing -- and usually outperforming -- AMD's own official open-source driver.

https://www.phoronix.com/vr.php?view=31245

[skeevy420]

That's not surprising. Except for this one time a couple years back with Hitman 2, with most every game I've played RADV has always performed better for me and AMDVLK usually had some minor issue.

What Michael's game benchmarks don't cover is shader processing induced issues that can occur while playing the game. Depending on the game there can be significant graphical issues until all the shaders are compiled. For example, when I played Snowrunner with AMDVLK 2022Q2.2 (the previous release) a couple weeks ago, turning on the headlights made the world and building textures disappear until all the shaders compiled. Took a couple of minutes. That's never happened once with RADV when starting from no shaders or starting with Valve's.

RADV, especially since ACO, seems to do a much, much better job with shader compiling than whatever AMDVLK does. As a gamer, I really don't get the point of AMDVLK anymore. I had high hopes back in the day, but trying random games I play on random AMDVLK releases over the years has lowered those hopes greatly.

[Quppa]

Good to see. On a non-gaming front, I'm still hopeful Mesa will get OpenCL image support in the not-too-distant future so we can stop waiting for Darktable/DaVinci Resolve/Blender to work with ROCm.

I know your time is limited, but I'd be really interested to see some like-for-like Windows 10 vs 11 vs modern Linux + Mesa (or proprietary drivers) + Proton gaming benchmarks, especially given that demand for the Steam Deck continues to exceed supply.

On a separate note, did the Enhanced Hardware Feedback Interface in Kernel 5.18 make any difference to Alder Lake performance, or do other parts of the software stack need to be updated to use it first?

[JacekJagosz]

RADV, especially since ACO, seems to do a much, much better job with shader compiling than whatever AMDVLK does. As a gamer, I really don't get the point of AMDVLK anymore. I had high hopes back in the day, but trying random games I play on random AMDVLK releases over the years has lowered those hopes greatly.

And the interesting thing is that AMD in their latest Windows driver seemed to move from their propietary shader compiler to LLVM (in this thread: https://forums.guru3d.com/threads/am...8#post-6023569 ). And on Linux that is by far the inferior option. So I think this comparison of RADV vs AMDVLK shows maybe their Windows driver could be much better too.

[M@GOid]

Nice to see that my new RX 6600 will be mostly trouble free the moment I plug it in the afternoon. After 2 agonizing years, I finally managed to grab a new GPU for a reasonable price.

[mahurinj]

Is there a reason for AMD to continue developing a separate open source driver rather than just contributing to the RADV driver and blessing that as the official supported version?

[dimko]

Nice to see that my new RX 6600 will be mostly trouble free the moment I plug it in the afternoon. After 2 agonizing years, I finally managed to grab a new GPU for a reasonable price.

Aamen. Albeit i was lucky, in between covid years i managed to grab 5700XT at MSRP.

[M@GOid]

Aamen. Albeit i was lucky, in between covid years i managed to grab 5700XT at MSRP.

Got mine in a flash sale. After a couple hours they where gone.

[tildearrow]

But with AMDVLK using the LLVM AMDGPU shader compiler rather than their in-house proprietary shader compiler

Imagine AMDVLK with ACO...

[bridgman]

Is there a reason for AMD to continue developing a separate open source driver rather than just contributing to the RADV driver and blessing that as the official supported version?

As far as I know we are not and have never developed a separate open source Vulkan driver - if we had it would look a lot like RADV although it probably would not have ACO yet.

What we are doing is taking the (multi-OS but primarily developed for Windows) closed source driver and periodically sanitizing a snapshot of the code required for a Linux build. The shader compiler code is different from the closed source version but is based on the same back end we use for open source graphics drivers and the ROCm stack.