Announcement

Collapse
No announcement yet.

Radeon+Ryzen CPUFreq CPU Scaling Governor Benchmarks On Linux 4.15

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

  • Radeon+Ryzen CPUFreq CPU Scaling Governor Benchmarks On Linux 4.15

    Phoronix: Radeon+Ryzen CPUFreq CPU Scaling Governor Benchmarks On Linux 4.15

    Taking a break from KPTI and Retpoline benchmarks, here are some tests recently conducted with Linux 4.15 when it comes to trying out the different CPUFreq scaling governors with this latest kernel and running various games with a Radeon RX 580 Polaris graphics card.

    http://www.phoronix.com/vr.php?view=25807

  • #2
    No surprises whatsoever, still, good to know.

    Comment


    • #3
      A slightly easier to remember command `sudo cpupower frequency-set -g performance` should also work for changing governor (change performance to whichever you want).

      These results pretty well echo what I've seen when playing Dota 2. Only with 'performance' do I get a mostly solid 60FPS (290, 1440p).

      Comment


      • #4
        The average FPS aren't the measure that benefits the most. It's the minimum FPS.

        In CSGO you can actually feel the difference. Here's watching a bot game, with a patch for adding a "low-fps" graph that is based on only the slowest frame time in the interval (https://gist.github.com/ChristophHaa...44440b5a922d):
        ondemand: https://i.imgur.com/FJ6rdwv.png
        performance: https://i.imgur.com/5lwFoPH.png
        To be honest, both results are REALLY terrible for an 1600X @ 3.9GHz + RX 480, csgo is just not really playable until it's ported to Source 2 + Vulkan.
        The frametimes are not very accurate but you can still see the difference it makes.

        The frame delay graphs (not mainline either) show how much frames were lagging behind the the 60/45 fps in the interval and they show it even better. It's interesting that the GPU load gets more "stable" on a lower level while the GPU clock goes a bit up. That suggests that with ondemand the GPU is not fed commands/data as reliably fast as with performance which could cause small stalls.

        The question is: What's the power usage if you just always run with the performance governor?
        Michael Can you measure power usage in idle and under load with the performance governor vs ondemand?
        Last edited by haagch; 01-07-2018, 10:49 AM.

        Comment


        • #5
          Tested it with The Witcher 3 in Wine (Ryzen 7 1700X, RX 480, 1920x1200). Not much of a difference really, but performance looks slightly more stable at around 40 fps.

          ondemand:



          performance:

          Last edited by shmerl; 01-07-2018, 11:00 AM.

          Comment


          • #6
            yeah its the fluctuations and drops. In my experience Powersave is the worst for gaming, but ofc it should be as its powersave. Ondemand is next bad in the list, fluctuates framerates so much, jumping up and down, clocks going up and down switching between cores, not smooth experience. Schedutil is a bit better, less fluctuations. Performance is the king, FPS is more stable and overall experience is super smooth, now that ofc depends on the game, minimum drops can still happen in some games.

            Ryzen 1700X 3.9ghz, GTX 1070 here, playing at 1440p.

            Comment


            • #7
              Wail... powersave sets the lowest freqency and so little difference. I expected much bigger difference, since games like higher cpu freq.

              Right?

              Comment


              • #8
                With proper mainboard uefi settings, using a OS governor to save power shouldn't be necessary anymore with modern CPUs.
                Some months ago I found out that setting Package C-State support for my Skylake to Auto (disabled by default, gets also disabled by some OC settings and thus has to be reenabled) also allows powersaving with OS governors set to maximum performance (CPUFreq Performance on Linux, maximum performance setting on Windows).
                I did check this with a wattmeter and the results regarding energy savings were as good as when using OS powersave governors (at least when fully [email protected]). And unlike OS powersave governors, I couldn't measure any performance drawbacks.
                All C-States had to be set to Enabled or Auto for this and Package C-State additionally to Auto as well.

                The same should also work for Ryzen and other CPUs, but it's probably up to a proper uefi which is not utterly bugged.
                Big advantage is no more worrying about any governor, except you shouldn't forget to turn off the powersaving ones since they still can hurt performance.

                Comment


                • #9
                  Originally posted by aufkrawall View Post
                  With proper mainboard uefi settings, using a OS governor to save power shouldn't be necessary anymore with modern CPUs.
                  Some months ago I found out that setting Package C-State support for my Skylake to Auto (disabled by default, gets also disabled by some OC settings and thus has to be reenabled) also allows powersaving with OS governors set to maximum performance (CPUFreq Performance on Linux, maximum performance setting on Windows).
                  I did check this with a wattmeter and the results regarding energy savings were as good as when using OS powersave governors (at least when fully [email protected]). And unlike OS powersave governors, I couldn't measure any performance drawbacks.
                  All C-States had to be set to Enabled or Auto for this and Package C-State additionally to Auto as well.

                  The same should also work for Ryzen and other CPUs, but it's probably up to a proper uefi which is not utterly bugged.
                  Big advantage is no more worrying about any governor, except you shouldn't forget to turn off the powersaving ones since they still can hurt performance.
                  Just Checked with my Ryzen. Schedutil governor idle 40-65W. Performance governor idle 70-90W. (thats the whole PC powerusage, without monitors)
                  Though i have C-states disabled since the freeze bug after some intel patch hit the 4.14 or 4.13 kernel which caused that. something to do with rcu_nocbs. Might try enabling c-states after kernel 4.15 hits stable mark, cause i heard its fixed there.

                  Comment


                  • #10
                    Originally posted by gsedej View Post
                    Wail... powersave sets the lowest freqency and so little difference. I expected much bigger difference, since games like higher cpu freq.

                    Right?
                    Games benefit from higher frequency when their performance is limited by the speed of a single CPU core. If performance is limited by something else, e.g. the GPU, then the speed of the CPU makes very little difference, hence powersave matching performance on some tests.

                    Note that this can be affected by other factors, such as the CPU IPC (Instructions Per Clock - doing more per clock cycle) or the ability of the game to use more CPU cores, reducing the dependency on the speed of each core.

                    Comment

                    Working...
                    X