Announcement

Collapse
No announcement yet.

AMD + Valve Focusing On P-State / CPPC Driver With Schedutil For Better Linux Efficiency

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • AMD + Valve Focusing On P-State / CPPC Driver With Schedutil For Better Linux Efficiency

    Phoronix: AMD + Valve Focusing On P-State / CPPC Driver With Schedutil For Better Linux Efficiency

    As reported at the start of August, AMD and Valve have been working on Linux CPU performance/frequency scaling improvements with the Steam Deck being one of the leading motivators. As speculated at that time, their work would likely revolve around use of ACPI CPPC found with Zen 2 CPUs and newer. Published last week was that AMD P-State driver for Linux systems indeed now leveraging CPPC information. AMD formally presented this new driver yesterday at XDC2021...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    The talk itself starts at https://youtu.be/UK4z42U1lSs?t=14127

    The track at the beginning of the clip: https://www.youtube.com/watch?v=0CCI8PePeH8
    Last edited by birdie; 18 September 2021, 08:09 AM.

    Comment


    • #3
      I wish AMD and Valve would sponsor GIMP as part as a game development toolkit fund.

      Comment


      • #4
        While CPPC sounds useful for portables and general OS use, I'm going to assume that using performance governors and setting higher priorities will likely work better for specific tasks we're aiming for maximum performance with. That'll be my assumption until benchmarks and real world usage show otherwise (I feel like I've been down a similar path before with Intel and Schedutil on my last PC).

        But I'm glad to see a Zen 2 article. It's a real tease to have a Zen 2 and read about all these new Zen 3 features.

        Michael, just curious, have you ever ran benchmarks with different nice levels? Back in my Q6600 days I'd run Firefox with 5 and games with -5 which had a noticeable impact on both FF and games pushing my hardware to the limits: choppy FF scrolling and videos; better FPS average in games. When you're rocking 1080p30 low/medium you gotta pull out every trick in the book.

        Comment


        • #5
          Originally posted by uid313 View Post
          I wish AMD and Valve would sponsor GIMP as part as a game development toolkit fund.
          Eh, Valve will use KDE on the Steamdeck, so it makes more sense for them to sponsor Krita (which is also better for creating artwork).

          Comment


          • #6
            Originally posted by skeevy420 View Post
            While CPPC sounds useful for portables and general OS use, I'm going to assume that using performance governors and setting higher priorities will likely work better for specific tasks we're aiming for maximum performance with. That'll be my assumption until benchmarks and real world usage show otherwise (I feel like I've been down a similar path before with Intel and Schedutil on my last PC).

            But I'm glad to see a Zen 2 article. It's a real tease to have a Zen 2 and read about all these new Zen 3 features.

            Michael, just curious, have you ever ran benchmarks with different nice levels? Back in my Q6600 days I'd run Firefox with 5 and games with -5 which had a noticeable impact on both FF and games pushing my hardware to the limits: choppy FF scrolling and videos; better FPS average in games. When you're rocking 1080p30 low/medium you gotta pull out every trick in the book.
            On custom desktops or on PCs with a screaming fan, maybe. But burst turbo is so aggressive in everything else these days that, in some workloads, you do want to save as much power/heat as possible to give the system more room to boost when needed.

            schedutil + turbo control, for instance, already helps my renoir laptop in a certain CPU + dGPU heavy ML workload, though that is a odd workload.
            Last edited by brucethemoose; 18 September 2021, 11:46 AM.

            Comment


            • #7
              Originally posted by brucethemoose View Post
              On custom desktops or on PCs with a screaming fan, maybe. But burst turbo is so aggressive in everything else these days that, in some workloads, you do want to save as much power/heat as possible to give the system more room to boost when needed.

              schedutil + turbo control, for instance, already helps my renoir laptop in a certain CPU + dGPU heavy ML workload, though that is a odd workload.
              +1

              Modern consumer processors can consume temporarily 250W-300W. This is never sustainable as you simply cannot get the heat out from the package (independent of cooler).
              For reference 250W-280W is what server processors can do in steady state and they have multiple times the cooling surface/capability.

              This is the reason why all-core boost is different from single-core boosts nowadays and is also the reason why the "performance governor" can nowadays hurt performance.
              It's not always evident in benchmarks especially when you hit all cores with heavy load anyway.

              Comment


              • #8
                Originally posted by Vistaus View Post

                Eh, Valve will use KDE on the Steamdeck, so it makes more sense for them to sponsor Krita (which is also better for creating artwork).
                Yeah, I would love to see AMD and Valve sponsor Krita too!

                Comment


                • #9
                  I think that if the Steam Deck is successful enough and Proton is perfected, they could have a desktop/laptop variant of SteamOS and essentially be an Arch-based beginner's distribution for Linux. Valve could succeed where Ubuntu and Fedora failed and be the first company to find a way to make consumer desktop Linux profitable and easy for beginners.

                  Comment


                  • #10
                    Originally posted by Prescience500 View Post
                    I think that if the Steam Deck is successful enough and Proton is perfected, they could have a desktop/laptop variant of SteamOS and essentially be an Arch-based beginner's distribution for Linux. Valve could succeed where Ubuntu and Fedora failed and be the first company to find a way to make consumer desktop Linux profitable and easy for beginners.
                    Not going to happen. The way to make consumer hardware easy is to make it an appliance the way Apple does. That goes against what Linux is. I don't see why Valve would want to take on the headaches of making a general purpose OS when they could just release a gaming appliance. There's no profit in it for them.

                    Comment

                    Working...
                    X