Valve developer Pierre-Loup Griffais mentioned on Twitter, about a new Mesa shader compiler for AMD graphics named "ACO" and they're calling for testers.
In the longer post on Steam, it goes over a brief history about Valve sponsoring work done by open-source graphics driver engineers, with it all being "very successful". The team has grown and they decided to go in a different direction with their work.
To paraphrase and keep it short and to the point, currently the OpenGL and Vulkan AMD drivers use a shader compiler that's part of the LLVM project. It's a huge project, it's not focused on gaming and it can cause issues. So, they started working on "ACO" with a focus on good results for shader generation in games and compile speed.
It's not yet finished, but the results are impressive as shown:
That is quite an impressive decrease in compile time! They expect to be able to improved that further eventually too, as it's currently only handling "pixel and compute shader stages". Valve also included some gaming results as well. Not quite as impressive when compared to the above perhaps, but every single bit of performance they can squeeze in is great:
With more detailed performance testing info available here. Now that it's looking pretty good, being stable in many games and seeing a reduction of stuttering they're looking for wider testing and feedback. Packages for Arch Linux should be ready later today, with Valve looking into a PPA for Ubuntu too. Interested in testing? See this forum post on Steam.
You can see the full post about it on Steam and more details on the Mesa-dev mailing list entry here. The code can also be viewed on GitHub.
This comes only recently after Valve released a statement about remaining "committed to supporting Linux as a gaming platform" as well as funding work on KWin. Really great to see all this!
Take a look at the benchmark results: the difference in error. And many ports were specially sharpened under RADV - this driver has advantage.
Valve would have been able to negotiate with AMD long ago and start developing AMDVLK together. :'(
Next build (around christmas), if AMD as something to offer comparable to the GTX 1660ti (performance, price and TDP), I might give another chance to AMD in another brand (Gigabyte or Sapphire).
Also I bet you're a little proud of having this site quoted in their post Mr. "The Community" :)
Last edited by Shmerl on 3 July 2019 at 6:39 pm UTC
Quoting: mphuZGod damn, why?
Did you read their post. They answer this question.
Quoting: KimyrielleSame. I got my current PC shortly before AMD became a solid option for GPUs again. I hate NVidia as a company and will gladly take my business to AMD next time. Won't happen too soon I guess. I tend to keep my PCs longer and longer these days.I'm in a similar position. I switched to NVidia around the start of the Linux Gaming Revolution ( :) ), and frankly I'm glad I did. It went against my open-source principles, but then so do most games, and AMD was hopeless back then. The open-source drivers were next to useless for anything other than desktop work (which, to be fair, had been fine until the games started coming), and Catalyst was an absolute mess. The way the NVidia proprietary driver Just Worked™ was a real breath of fresh air.
Now though, if I was in the market for a new GPU (which, sadly, I'm not; I managed to score a GTX960 for free a few months ago, and it'll have to do me for a while yet), I'd definitely be looking at AMD.
See more from me