No announcement yet.

Benchmarking Five Linux Distros Against Windows 11 On The Threadripper PRO 7995WX / HP Z6 G5 A

  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Nice to see the next version of Debian being tested too and being the second in the ranking!


    • #32
      Originally posted by unwind-protect View Post
      How, exactly, does Windows manage to be so slow?
      It is a combination of several factors. For example, Windows does not have a modern filesystem and uses on-access antivirus scanning, which seriously degrades I/O performance compared to Linux.
      In tests where Windows performs similarly on few core CPUs and loses badly on many core CPUs, the culprit is often the scheduling. Threadripper PRO 7995WX has 96C/192T which exceeds the size of a single Win32 processor group, and requires for full utilization that the application is aware of processor groups and how to move threads between them.

      In some but not all cases, the Windows Server and Enterprise editions can work around this problem by deciding whether to assign a non-default processor groups at process creation time.
      Originally posted by unwind-protect View Post
      I mean a job like image denoise should be entirely CPU driven, no?
      Image denoise is actually one of the benchmarks where Windows 10 Pro has a scaling problem that does not exist with Windows 10 Enterprise. Not sure if Windows 11 Enterprise would have made a difference.

      The above image is from the TR 3990X thread scaling benchmarks, hopefully Michael will do thread scaling benchmarks on the 7995WX too.

      Originally posted by HD7950 View Post
      Those who wonder why Windows runs so slowly are still not familiar with VBS, HVCI and KDP. I wish we had something similar in Linux.
      VBS and HVCI hardly affect CPU bound tasks like Blender, especially on more modern CPUs. And Linux is faster in Blender by a considerable margin.

      Source: Tom's Hardware


      • #33
        Originally posted by Volta View Post
        Windows and security are opposite terms. It seems you forgot Linux has SELinux and AppArmor which impacts performance. It's like testing marketing BS with reality. Reality has won. Oh, and Linux has proper ASLR implementations and I doubt Windows comes even close to this.
        Whether or not you want to believe it, Windows 10 and 11 have excellent security.

        Windows has had DEP and ASLR since Win XP sp3.

        VBS on Win 10/11 can slow down performance between 5% and 15%:

        Of the distros tested, only Ubuntu uses AppArmor and only CentOS Stream uses SELinux.

        Try again.


        • #34
          Originally posted by agd5f View Post

          VBS is mostly just using the IOMMU to isolate device DMA. Linux has had this enabled for probably a decade before windows did.
          It really is a shame when a person that writes code for a major computer component manufacturer spreads FUD:

          Virtualization-Based Security (VBS) uses Windows Hypervisor to virtually isolate a segment of main memory from the rest of the operating system. Windows uses this isolated, secure region of memory to store important security solutions like log-in credentials and code responsible for Windows security, among other things.


          • #35
            Originally posted by andyprough View Post
            Wow, Windows gets destroyed once again, who could have possibly predicted that?

            It's almost as though MS, like Apple, is more than happy to cripple the performance on what should otherwise be fast hardware.
            You're right. I was actually in a top secret Microsoft meeting and the discussion was how MS could screw with their customers by purposely making it run slower than it has to.

            Who tipped you off?


            • #36
              Originally posted by sophisticles View Post
              There's no FUD. Until VBS, Windows didn't even enable the IOMMU on bare metal. There was no device IO isolation. Linux has enabled this for years. It's the reason the DMA API exists on Linux. There is more to VBS than just the IOMMU, but that is a big part of it.


              • #37
                Originally posted by Volta View Post

                Windows has won once again by a large margin.. in number of last place finishes! You can now crawl out from under the rock.
                Serious question, without Googling, do you know what most of the software tested is used for?

                When was the last time you used any of these:

                JPEG XL libjxl
                WebP Image Encode
                Intel Open Image Denoise
                Chaos Gtoup V-RAY​

                I swear, the AMD faithful are like some guy getting excited because car A is faster than car B in The Race to the Clouds​, even though they don't live anywhere near Colorado.


                • #38
                  Originally posted by sophisticles View Post
                  I was actually in a top secret Microsoft meeting
                  So Microsoft has special meetings for their one fan that cheerleads for them on GNU/Linux forums? Who knew?

                  You should demand more money. Whatever they are paying you, it's not nearly enough for all the work you do.


                  • #39
                    This looks pretty compelling for Tumbleweed. It ties for 2nd in first place finishes, and would be 2nd outright out of the box. It also ties for 1st with the fewest last place finishes. And it's doing so on a COW filesystem with automatic snapshots / rollback built in.


                    • #40
                      Considering its recent release, it would've been cool to see FreeBSD 14's results.