Announcement

Collapse
No announcement yet.

Windows 10 vs. Ubuntu 21.04 On The AMD Ryzen Threadripper 3990X

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

  • Windows 10 vs. Ubuntu 21.04 On The AMD Ryzen Threadripper 3990X

    Phoronix: Windows 10 vs. Ubuntu 21.04 On The AMD Ryzen Threadripper 3990X

    Earlier this month were benchmarks looking at Windows 10 vs. Ubuntu 21.04 on an AMD Ryzen 9 5900X desktop to which Ubuntu came out roughly 8% faster than the Microsoft OS on average. But what about the difference for HEDT systems? Given the more radical performance difference we have seen in the past with Windows vs. Linux for Threadripper systems, here are some recently conducted benchmarks on that front with the 64-core Threadripper 3990X.

    https://www.phoronix.com/vr.php?view=30217

  • #2
    Those are some substantial gains in that Windows update. Pretty weird how DAV1D is faster on Windows.

    Comment


    • #3
      It's been like this for a while though. Funny is that it's only the 1080p test that looks pretty bad. I also wonder what's up with JtR and OpenMP. It doesn't seem to matter what hash you use, it simply doesn't scale above more than 4 threads.

      Comment


      • #4
        I want just to point that the latest stable release of Windows is 21H1 19043 released ~ a week ago.

        Comment


        • #5
          Originally posted by Setif View Post
          I want just to point that the latest stable release of Windows is 21H1 19043 released ~ a week ago.
          Yeah these tests were started before that, given the time it takes to run all the tests and then to get it into article queue, but from some quick follow up tests I did on 19043 no real change in performance either way.
          Michael Larabel
          http://www.michaellarabel.com/

          Comment


          • #6
            Really goes to show how much better on average Linux's dynamic CPU governor (schedutil) performs compared to Windows' default which they call "Balanced".

            And I'm pretty sure things will only get better in Linux-land...

            BTW, does Microsoft actually still work on their kernel at all?
            I mean, it has been at version 10.0 since 2015!

            Comment


            • #7
              So choose application specific OS?
              Not ASIC but ASOS 🤔

              Comment


              • #8
                Originally posted by Linuxxx View Post
                Really goes to show how much better on average Linux's dynamic CPU governor (schedutil) performs compared to Windows' default which they call "Balanced".
                No, all it shows is that it doesn't make much sense to benchmark software between Ubuntu and Windows, where the Windows version is just an afterthought. For instance, John the Ripper is built on Windows using Cygwin. You can't even call it a port. So at this point, you might as well benchmark Excel on Windows vs Excel running via Wine on Ubuntu. It's utterly useless and I have no idea why Michael even includes such "benchmarks" in these cross-platform comparisons.

                99% of the time it boils down to the amount of effort put into a specific port of given software. If you think that either kernel is inherently superior to the other, you're a fanboy, there's no other way to put it.

                Originally posted by Linuxxx View Post
                BTW, does Microsoft actually still work on their kernel at all?
                I mean, it has been at version 10.0 since 2015!
                Not sure if trolling or just dumb.

                Comment


                • #9
                  Originally posted by anarki2 View Post
                  99% of the time it boils down to the amount of effort put into a specific port of given software. If you think that either kernel is inherently superior to the other, you're a fanboy, there's no other way to put it.
                  That is not really true. The items like blender in that benchmark where windows is losing its not the amount of effort put into a specific port that is the problem. Please note you also see with exactly the same binary big improvements in performance on Windows.

                  The number of developers with blender work on the Windows port is greater than the Linux port yet the performance is still bad for windows.

                  The Linux kernel is superior in particular places lower syscall cost than Windows kernel is quite a big one. There are places you see windows unfair scheduler work out beneficial in the benchmarks.

                  Majority of the items in Phoronix Test Suite are fair development.

                  Please do also be aware anarki2 you said benchmarking with Excel running under wine. I am sorry to have to break this to you there are particular Excel benchmarks running under wine that are faster than running the same benchmark on Windows. Please note those are all Excel benchmarks that work without Excel crashing due to something not being supported by wine. I am not saying the Linux kernel is perfect. Horrible reality here is you do get a better showing with wine under Linux than you do with cygwin on windows by quite a large margin to the point the wine option can be the winner due to the Linux kernel differences.

                  The reason why wine is not used in the Phoronix test suit is more wine performance is very dependant on what the application users and the wine developers have asked it not to be used because it would be deceptive to end users. Some applications can be up to 200% faster than windows others can be less than 10% the speed of windows with wine. Yes where you are 200% faster this is because Linux and wine can do something a lot faster than windows. Yes that 200% faster is no joke and is not a typo.

                  Cygwin is a lot more predictable beast than wine for amount of overhead.

                  The reality here is a port to Linux that is done to the grade a port to Cygwin is the other way generally performs better than jack the ripper is showing. They are normally inside 90% on Linux of their windows version. Normally if they are not inside 90% the companies normally don't release them for Linux so only provide a windows binary forcing the use of wine to run the programs.

                  So there is a question that Jack the Ripper raises does Cygwin need more optimisation investment to bring the Cygwin overhead into line with wine?

                  Comment


                  • #10
                    Originally posted by oiaohm View Post
                    The Linux kernel is superior in particular places lower syscall cost than Windows kernel is quite a big one. There are places you see windows unfair scheduler work out beneficial in the benchmarks.
                    I think I remember seeing Windows developer talk about how WSL1 didn't work out because Linux apps expect read/write to the filesystem to be blazing fast while NTFS and the Windows kernel simply couldn't keep up.

                    Linux certainly isn't perfect, but trying to beat it in a drag race is a recipe for failure.

                    Comment

                    Working...
                    X