Announcement

Collapse
No announcement yet.

Windows vs. Linux Scaling Performance From 16 To 128 Threads With AMD Ryzen Threadripper 3990X

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

  • Windows vs. Linux Scaling Performance From 16 To 128 Threads With AMD Ryzen Threadripper 3990X

    Phoronix: Windows vs. Linux Scaling Performance From 16 To 128 Threads With AMD Ryzen Threadripper 3990X

    As has been known for a while now, AMD Ryzen Threadripper processors really show their true potential on Linux with often significant increases to the performance thanks to the kernel's better scalability compared to Microsoft Windows. While Microsoft has made some improvements in this area over the past year, with the AMD Ryzen Threadripper 3990X 64-core / 128-thread HEDT processor it really shines on Linux. In this article are benchmarks of Windows 10 Professional and Windows 10 Enterprise against Linux on the Threadripper 3990X when going from 16 cores to 128 threads for seeing how the three operating systems are scaling.

    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

  • #2
    Typo:

    Originally posted by phoronix View Post
    Windows 10 Pro Build 18363 (Novemner 2019 Update).

    Comment


    • #3
      I would like to see a comparison between AMD Ryzen 3950X and Threadripper 3990X set at 16 cores and compare power consumption in Clear Linux. I don't care about Windows anyway. Looking at benchmarks for Blender, 32 cores seem like a nice performance increase.

      Hopefully software programs will get better with CPU core+SMT scaling over time.

      Comment


      • #4
        These results coincide with other test sites. Microsoft is pushing the scheduler update through the Enterprise release, not the Pro. Word on the street is that Pro will be discontinued soon as a SKU.

        For those who want Epyc instead of TR, many resellers are starting to move Gen 1 Epyc rack servers out the door on the cheap.

        Comment


        • #5
          Shame apps don't scale. How would these benchmarks do if run multiple concurrent runs?

          when i do ffmpeg on my ryzen it uses about 60% CPU only, even when fully threaded. So running two ffmpegs in parallel gives me almost double throughput.

          Comment


          • #6
            Originally posted by Spam View Post
            Shame apps don't scale. How would these benchmarks do if run multiple concurrent runs?
            I think these CPUs are more built for multitasking rather than massive parallelization for a single application. There's only so far a program can scale with more cores.

            Comment


            • #7
              I had to sign into a Windows 10 system today to check on something - looks like a desktop from the 1990's compared to a modern distro running Plasma. When was the last time the file manager saw an upgrade? Windows Vista?

              Comment


              • #8
                Very interesting article, thanks Michael.
                ## VGA ##
                AMD: X1950XTX, HD3870, HD5870
                Intel: GMA45, HD3000 (Core i5 2500K)

                Comment


                • #9
                  In the early 2000's, I worked at a small company that supplied mapping data to Google and others. They flew small planes with massive cameras mounted on them, where the resulting images were something like 60 Megapixels, which isn't that impressive today, but was insane back in ~2002. My job there was maintaining the computer they used to stitch all the images together and correlate GPS coordinates and street name/location with the imaging data. It was an SGI Origin 2800 with 128 CPU's and IIRC we had upgraded it to 256 GB of memory just before I left. The dollar cost of this SGI server was in the $10 Million range at the time. It's truly amazing that in under 20 years, we now have the same computing capability in a (very handsome looking!) desktop tower costing a few thousand bucks.
                  Last edited by torsionbar28; 14 February 2020, 03:13 PM.

                  Comment


                  • #10
                    Originally posted by torsionbar28 View Post
                    In the early 2000's, I worked at a small company that supplied mapping data to Google and others. They flew small planes with massive cameras mounted on them, where the resulting images were something like 60 Megapixels, which isn't that impressive today, but was insane back in ~2002. My job there was maintaining the computer they used to stitch all the images together and correlate GPS coordinates and street name/location with the imaging data. It was an SGI Origin 2800 with 128 CPU's and IIRC we had upgraded it to 256 GB of memory just before I left. The dollar cost of this SGI server was in the $10 Million range at the time. It's truly amazing that in under 20 years, we now have the same computing capability in a (very handsome looking!) desktop tower costing a few thousand bucks.
                    Well, it has always been like this.... Once upon a time a whole hotel room of cpus had less processing power than a modern mid range phone.... Imagine what kind of beast you can buy for 10m in 2020.....

                    Comment

                    Working...
                    X