Announcement

Collapse
No announcement yet.

AMD Ryzen 7 5800X3D - Windows 11 vs. Ubuntu 22.04 LTS Linux Benchmarks

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

  • #21
    Originally posted by nranger View Post

    The cpufreq governor only applies to the CPU cores, while the SteamDeck runs on an SoC that shares power with the built in GPU, and the whole thing is generally power-limited anyway. If the game you're playing is GPU limited, then having the governor push the cpu to it's highest state at all times will really only hurt framerates, or at best, burn unnecessary power.
    That's a common misconception, as can be seen by these benchmarks where Clear Linux with AMD-PSTATE + performance governor is not starving the iGPU of the Ryzen 7 PRO 5850U at all; in fact, quite the contrary:

    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


    In Michael's own words:

    Even with Clear Linux using the performance governor by default rather than Schedutil, it didn't equate to any measurably higher CPU power consumption or CPU core temperature for the HP Dev One during this Linux gaming testing.

    Comment


    • #22
      Originally posted by Volta View Post

      Maybe they were aiming at lower power consumption, but making some games unplayable or stuttering isn't worth it. Or perhaps, this governor is not broken for every CPU?
      I doubt it.

      From my experience, any governor that tries to be "smart" about the choices it makes, will inevitably lead to stuttering during gameplay.

      Also, it doesn't matter how powerful your CPU is; take the following example:

      "Destroy All Humans! 2 - Reprobed" is an unoptimized Unreal Engine 4 title which was recently released.

      A user on ProtonDB with an Intel Core i9-9900 + NVIDIA GeForce RTX 2080 fixed the game on Linux with the following:

      Note: Without Proton GE or GameMode I experienced stuttering.
      So basically the performance governor came to the rescue once again...

      Comment


      • #23
        Originally posted by birdie View Post
        I wonder if Michael has enabled any of the optimizations (a list for W10) required to carry out benchmarks on Windows.
        Windows runs a ton of random crap in background which may roally skew the results.
        i did pay michael in the past to make w10privacy tool benchmarks... https://www.w10privacy.de/

        in some benchmarks it helps a lot.

        but face it microsoft windows is in fact broken by design and microsoft will never fix it if you make money by this broken by design and if you fix it the money flow to you stops you will never fix it.
        Phantom circuit Sequence Reducer Dyslexia

        Comment


        • #24
          Originally posted by Michael View Post
          Unfortunately not possible for lack of PHP PCNTL extension on Windows for being able to concurrently and accurately poll the power data while running benchmarks.
          can you redo the benchmark of w10privacy tool on windows 11 ? https://www.w10privacy.de/
          Phantom circuit Sequence Reducer Dyslexia

          Comment


          • #25
            Originally posted by Michael View Post

            Unfortunately not possible for lack of PHP PCNTL extension on Windows for being able to concurrently and accurately poll the power data while running benchmarks.
            How about continuous measurement of total power consumed during the benchmarks run, measured at the wall plug together with the start and end times for each benchmark?

            Power supplies should provide power consumption data to the system. Power stats would be helpful identifying rogue processes (malware) when you see high usage a times the system is normally idle.

            Comment


            • #26
              Originally posted by gnwiii View Post

              How about continuous measurement of total power consumed during the benchmarks run, measured at the wall plug together with the start and end times for each benchmark?

              Power supplies should provide power consumption data to the system. Power stats would be helpful identifying rogue processes (malware) when you see high usage a times the system is normally idle.
              PTS needs the PHP PCNTL extensions to be able to concurrently log the system power consumption data in real-time for the entire duration of the run as well as coordinating the per-test power/thermal results. But PHP PCNTL simply doesn't exist on Windows and there isn't any similar process/threading interface from PHP that I know of that nicely works on Windows. So for lack of time/resources to come up with some other solution, just not possible.
              Michael Larabel
              https://www.michaellarabel.com/

              Comment


              • #27
                Are there any reasons for x265 being replaced by kvazaar in HEVC encoding test? is it a matter of licensing? or quality?

                Comment


                • #28
                  Originally posted by Linuxxx View Post

                  The only data we have on that are the only benchmarks Michael did on the Steam Deck during the launch window of the device, which clearly shows "acpi-cpufreq schedutil" is the default:

                  https://www.phoronix.com/review/stea...teamos-linux/4

                  You might also remember the following announcement months prior to the launch of the Steam Deck:

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

                  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


                  Looks like they didn't realize back then just how broken schedutil really is when it can freely choose any frequency it wants, instead of having only 3 different frequencies to choose from as is the case with acpi-cpufreq.​
                  but that was a beta state of amd-pstate. and the kernel in the first link is 5.13 apparently. amd-pstate was release in 5.17 i think

                  Comment


                  • #29
                    Originally posted by topolinik View Post
                    Are there any reasons for x265 being replaced by kvazaar in HEVC encoding test? is it a matter of licensing? or quality?
                    Oh I never caught that, I did not know about that HEVC encoder I wonder if I should use it instead of x265.

                    Comment


                    • #30
                      Originally posted by loganj View Post

                      but that was a beta state of amd-pstate. and the kernel in the first link is 5.13 apparently. amd-pstate was release in 5.17 i think
                      But do note that this a custom kernel by Valve, which for example already carried the Fsync patches, whereas they were first upstreamed to Linux 5.16.

                      So if AMD-PSTATE were to be an improvement, Valve would have certainly backported it already.

                      Alas, doesn't look like it was deemed worthwhile by them...

                      Comment

                      Working...
                      X