Announcement

Collapse
No announcement yet.

Clear Linux Offers Up Advantages For Ice Lake Xeon, CentOS Comes In Strong

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

  • Clear Linux Offers Up Advantages For Ice Lake Xeon, CentOS Comes In Strong

    Phoronix: Clear Linux Offers Up Advantages For Ice Lake Xeon, CentOS Comes In Strong

    Earlier this week when posting Ubuntu 20.04 LTS / 20.10 / 21.04 benchmarks on the new Intel Xeon Platinum 8380 "Ice Lake" server processors, one of the first questions that came up was about how well these new 10nm server CPUs perform with Intel's own Clear Linux distribution. While Clear Linux releases have become much less frequent and far less to communicate these days on new improvements/optimizations among other ongoing shifts with that Intel open-source project, it is still performing very strongly with 3rd Gen Xeon Scalable hardware. CentOS in these tests also had a strong showing with the increasing performance focus on that front.

    https://www.phoronix.com/vr.php?view=30205

  • #2
    Really great to see the performance gap is narrowing - assuming Clear isn't regressing, it shows that these other distros are finally taking advantage of the same sorts of optimizations.

    Comment


    • #3
      God, the latest Intel chips seem to be having quite some issues with the powersave governor. I recall that even RKL was noticeably slower with the powersave governor. Michael , any chance you could run a few tests on, say, Ubuntu 21.04 with performance governor and see if it closes the gap?

      Comment


      • #4
        Originally posted by MadCatX View Post
        God, the latest Intel chips seem to be having quite some issues with the powersave governor. I recall that even RKL was noticeably slower with the powersave governor. Michael , any chance you could run a few tests on, say, Ubuntu 21.04 with performance governor and see if it closes the gap?
        As said in the article, there are some P-State tests happening at the moment (on Ubuntu 21.04 with Linux 5.13) with perf/power that should provide more insight.
        Michael Larabel
        http://www.michaellarabel.com/

        Comment


        • #5
          1. I'm eager to see tests which are interesting to a whole lot more people. For instance, let's compare Fedora vs Ubuntu vs Windows 10 in terms of
          • Raw performance (web browsing, office suites, rendering, encoding, compiling)
          • Battery life when running light tasks or watching YouTube
          • Idle power consumption (yeah, it can be wildly different from system to system)
          • Games, only not those which offer 800fps on modern PCs, but those which are heavy on shaders.
          and not just on a single system, but many, since e.g. 10th/11th gen Intel CPUs have DPTF under Windows and none under Linux.

          2. What about comparing e.g. Chrome/Firefox under Linux to their versions under Windows on a desktop system?

          3. What about comparing the performance of games running under Windows to them running under DXVK since we don't have anything better?

          In 2021 alone I've seen at least a dozen tests of ClearLinux with lots of server oriented tasks - how many people out there use it? A dozen? For how many people are those tests? What's the point of all these comparisons if you don't try to find out what makes this or that distro faster and you don't file bug reports to make other distros faster as well? Is it the kernel version? Kernel features? GCC version or/and compilation flags?

          You're doing a ton of work and sometimes it feels like there's very little output.

          Comment


          • #6
            Half of distros are based on Debian and Debian is nowhere to be seen and it's been for a while. So much emphasis on dead distro like CentOS (I am not tlaking about CentOS Stream).

            Comment


            • #7
              Originally posted by birdie View Post
              1. I'm eager to see tests which are interesting to a whole lot more people. For instance, let's compare Fedora vs Ubuntu vs Windows 10 in terms of
              • Raw performance (web browsing, office suites, rendering, encoding, compiling)
              • Battery life when running light tasks or watching YouTube
              • Idle power consumption (yeah, it can be wildly different from system to system)
              • Games, only not those which offer 800fps on modern PCs, but those which are heavy on shaders.
              and not just on a single system, but many, since e.g. 10th/11th gen Intel CPUs have DPTF under Windows and none under Linux.

              2. What about comparing e.g. Chrome/Firefox under Linux to their versions under Windows on a desktop system?

              3. What about comparing the performance of games running under Windows to them running under DXVK since we don't have anything better?

              In 2021 alone I've seen at least a dozen tests of ClearLinux with lots of server oriented tasks - how many people out there use it? A dozen? For how many people are those tests? What's the point of all these comparisons if you don't try to find out what makes this or that distro faster and you don't file bug reports to make other distros faster as well? Is it the kernel version? Kernel features? GCC version or/and compilation flags?

              You're doing a ton of work and sometimes it feels like there's very little output.
              What would you consider a rational representation of desktop users and their hardware for these tasks?
              Same question for laptops when testing battery life?
              What games do you think represent the qualification of "heavy on shaders".

              Correct me if I am wrong but every laptop Phoronix has tested includes a Chrome/Firefox test, can't you simply pull a report from openbenchmarking.org and look at the results?

              The instructions on how to set up Wine for DXVK are found here:

              https://linuxconfig.org/improve-your...inux-with-dxvk

              Pick a game to install. Anything that runs on DirectX 11 is a good candidate to test out. Keep in mind that not every game runs better with DXVK. It's still a very young project, and it's not optimized for every situation yet. This guide is going to follow Overwatch. It's a fairly popular DX11-only game, and it works well with Lutris.

              I would be interested in knowing which games meet the criteria.

              Comment


              • #8
                Originally posted by birdie View Post
                3. What about comparing the performance of games running under Windows to them running under DXVK since we don't have anything better?
                Actually, quite a few AMD GPU owners running Windows have discovered that using DXVK improves their gaming experience quite significantly.

                Comment


                • #9
                  Regarding power usage, a little while back there was a windows AMD GPU driver update that conserved quite a bit of power by not ramping up every component of the silicon when 3d hardware was in use (ostensibly the assumption was 3d being activated was for a heavy workload, so everything else needed to turn on/clock up as well.) GLAMOR is used heavily on Linux to provide 2d acceleration by performing it over 3d hardware, needlessly consuming extra power. I wonder if there's a similar situation with other hardware, with a similar solution.

                  Comment


                  • #10
                    Originally posted by Michael View Post

                    As said in the article, there are some P-State tests happening at the moment (on Ubuntu 21.04 with Linux 5.13) with perf/power that should provide more insight.
                    On my side, with Ubuntu, I never saw much difference between "On Demand" and "Performance" using custom benchmark or Phoronix benchmark. I never tried the other CPU governor options.

                    Comment

                    Working...
                    X