Announcement

Collapse
No announcement yet.

Windows 10 Outperforming Linux On A ~$5000 Laptop, Ubuntu Beating Clear Linux

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

  • #41
    Originally posted by hansdegoede View Post
    This probably is a thermal management issue, where the BIOS is configuring the CPU for a low TPD and relying on Windows to dynamically update the TPD depending on various factors including e.g. accelerometer readings which are used to detect if the laptop is on a lap (underside must not get too hot) or on a table (underside may get somewhat hotter).

    Also see: https://github.com/erpalma/throttled

    Michael, it would be good if you can run one of the benchmarks with a considerable difference between Windows and Linux on battery, while monitoring the power-consumption from the battery with e.g. powertop under Linux and something similar under Windows. Or you could look at the wall-outlet power when fully charged, so the only power consumed is the one directly used for processing. I would not be surprised if under Windows significantly more power is being drawn.

    What might also be fun is putting the laptop on your lap while running a (short running) benchmark under Windows and moving your knees up and down a bit (to make sure it detects it is on a lap) this might reduce the Windows benchmark results ...
    i feel like this is the issue as well. considering the round of benchmarks done a few days ago with ubuntu vs clear linux i really feel like it was some sort of thermal throttling or boosting issue under linux.

    Comment


    • #42
      Originally posted by MadCatX View Post
      What were the CPU clocks like under load? This smells like incorrectly set TDP limit under Linux.
      That does seem like the obvious answer, it's just curious it effects Clear Linux more than the other distros.

      Comment


      • #43
        Originally posted by smitty3268 View Post
        That does seem like the obvious answer, it's just curious it effects Clear Linux more than the other distros.
        If I had to guess, I'd say that it is due to Clear's aggressive use of AVX2 and AVX512. Powering the whole YMM/ZMM circuitry uses a lot of power and AFAIK even desktop CPUs need to throttle down when running some AVX2+-heavy code.

        Comment


        • #44
          Originally posted by MadCatX View Post
          If I had to guess, I'd say that it is due to Clear's aggressive use of AVX2 and AVX512. Powering the whole YMM/ZMM circuitry uses a lot of power and AFAIK even desktop CPUs need to throttle down when running some AVX2+-heavy code.
          On top of that, it could be pinning tasks/threads to individual cores more. Causing more heat and therefore more throttling. I don't know exactly what Clear Linux does, never ran it and not much need for it.

          Comment


          • #45
            I've been using System76 laptops for a few years, currently on a GalagoPro (galp3-pro) and it runs really, really fast with Ubuntu 18.04.4 LTS

            Comment


            • #46
              Well this article ruined my yesterday. Spent the whole day for installing exactly same Windows version, build. Hoping that there is really something changed. But... I ran appleseed test under fresh Windows and fedora 31. Here is a results
              WINDOWS
              Scene: Emily (Seconds ↓ )159.055
              Scene: Disney Material (Seconds ↓ )64.309
              Scene: Material Tester (Seconds ↓ )169.736
              LINUX

              Scene: Emily (Seconds ↓ )137.16
              Scene: Disney Material (Seconds ↓ )62.40
              Scene: Material Tester (Seconds ↓ )137.61

              I did that because I'm 3D artist and use Maya with Arnold renderer. Under linux Arnold is for 30% faster then Windows. It is significant time saving when you rendering animation.
              There is definitely something wrong with that laptop.

              Comment


              • #47
                Another interesting & amazing test, with unexpected results.
                1) Is Clear Linux open source? If they only have a small development team, in house, this might show the superiority of a large open source development team.
                2) Many versions (4) of Firefox were used. The worst result used the oldest version. The versions of the Linux kernel should have been different as well. Did these bias the results?
                3) Real world use of the operating system usually involves malware & virus protection. In this testing environment, Microsoft Windows probably did the VW-trick. When detecting bench testing, it turned off the malware protection, and increased the many caches and cache speeds.
                4) All operating systems, except two (Microsoft & Clear) are general usage systems. The correct general usage Microsoft system is NOT the one used in this series of tests.
                5) NVIDIA claims that this year it will better assist its drivers for Linux. We can assume that the drivers for Windows are already very well optimised.
                6) Manjaro is based from Arch, but proudly avoids the Arch direct Arch repositories, preferring its own versions of safety. Are the Arch enthusiasts justified in their superiority claims?
                7) Most single users of operating systems use operating systems designed for single users; not as servers nor as workstations. Is there a benchmark difference here?
                8) As usual, there is the unanswered question. Did Clear Linux use the oldest Linux kernel? Are later versions of the kernel faster?
                9) The hardware units are the most recent in model & manufacture. Generally most end users do not have this very recent hardware. The system creators most linked with hardware advances should be best to take advantage of this hardware.
                10) Bench tests should mention environmental conditions. A room set at 20, 30 or 40 degrees centigrade will give very different results. High humidity & booster fans might also lead to better heat transfer.
                11) Phoronix is a reputable tester. Great care should be given when any hardware or software seller claims better comparison results.

                Comment


                • #48
                  Very glad to see HP sending over a laptop for review. Also because it definitely shows a big problem. I hope it will be fixed sooner rather than later. Until then this is a no-go, for such money this kind of problem on an high performance machine is a killer.

                  Comment


                  • #49
                    Originally posted by gregzeng View Post
                    Another interesting & amazing test, with unexpected results.
                    1) Is Clear Linux open source? If they only have a small development team, in house, this might show the superiority of a large open source development team.
                    2) Many versions (4) of Firefox were used. The worst result used the oldest version. The versions of the Linux kernel should have been different as well. Did these bias the results?
                    3) Real world use of the operating system usually involves malware & virus protection. In this testing environment, Microsoft Windows probably did the VW-trick. When detecting bench testing, it turned off the malware protection, and increased the many caches and cache speeds.
                    4) All operating systems, except two (Microsoft & Clear) are general usage systems. The correct general usage Microsoft system is NOT the one used in this series of tests.
                    5) NVIDIA claims that this year it will better assist its drivers for Linux. We can assume that the drivers for Windows are already very well optimised.
                    6) Manjaro is based from Arch, but proudly avoids the Arch direct Arch repositories, preferring its own versions of safety. Are the Arch enthusiasts justified in their superiority claims?
                    7) Most single users of operating systems use operating systems designed for single users; not as servers nor as workstations. Is there a benchmark difference here?
                    8) As usual, there is the unanswered question. Did Clear Linux use the oldest Linux kernel? Are later versions of the kernel faster?
                    9) The hardware units are the most recent in model & manufacture. Generally most end users do not have this very recent hardware. The system creators most linked with hardware advances should be best to take advantage of this hardware.
                    10) Bench tests should mention environmental conditions. A room set at 20, 30 or 40 degrees centigrade will give very different results. High humidity & booster fans might also lead to better heat transfer.
                    11) Phoronix is a reputable tester. Great care should be given when any hardware or software seller claims better comparison results.
                    As a habit Phoronix does not "normalize" the builds prior to testing, he downloads their latest release as is and tests. Occasionally on smaller head to head testing he will normalize the kernel to see if/what changes that kernel is bringing overall.

                    If the default release from a distro has an old Firefox, that is what he will test, since that is what one would get "out of the box".

                    I am not sure what your "general usage OS" comment is meant to convey.

                    Clear Linux is a rolling release distro.

                    I believe some other people have expressed environmental condition questions (like on a real lap, or sitting on a table, etc).

                    Either way, Michael has committed to a follow up set of tests. It will be interesting to see what is happening here.

                    Comment


                    • #50
                      It's at times like this I wish I'd kept myself up to date with machine code/assembly language rather than just leaving it to younger people.

                      Ah well.

                      I've found with my edge case, Clear Linux smoking Windows 10 but YMMV.

                      Comment

                      Working...
                      X