Announcement

Collapse
No announcement yet.

Linux 5.6 Tests On AMD EPYC 7742 vs. Intel Xeon 8280 2P With 100+ Benchmarks

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

  • Linux 5.6 Tests On AMD EPYC 7742 vs. Intel Xeon 8280 2P With 100+ Benchmarks

    Phoronix: Linux 5.6 Tests On AMD EPYC 7742 vs. Intel Xeon 8280 2P With 100+ Benchmarks

    The latest benchmarks for your viewing pleasure are looking at the dual Intel Xeon Platinum 8280 performance up against the dual AMD EPYC 7742 CPUs while using the in-development Linux 5.6 kernel as the first time trying out these highest-end server processors on this new kernel debuting as stable in about one month's time...

    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

    Comment


    • #3
      Michael feedback of the PDF results:


      I like one line for sentence/information, so my feedback might be biased

      I would change the part "The results with the greatest spread from best to worst included" into bullet lists or a table
      • Tungsten Renderer (Scene: Volumetric Caustic) at3.53x,
      • MBW (Test: Memory Copy - Array Size: 4096 MiB) at 3.22x,
      • Parboil (Test: OpenMP MRI Gridding) at 3.02x,
      • Hackbench (Count: 32 - Type: Process) at 3.01x,
      • Tachyon (Total Time) at 2.62x
      • C-Ray (Total Time - 4K, 16 Rays PerPixel) at 2.37x
      • dav1d (Video Input: Chimera 1080p) at 2.33 (.....)

      Same for "Test Systems":
      • Processor: 2 x Intel Xeon Platinum 8280 @ 4.00GHz (56 Cores / 112 Threads)
      • Motherboard: GIGABYTEMD61-SC2-00 v01000100 (T15 BIOS)
      • Chipset: Intel Sky Lake-E DMI3 Registers
      • Memory: 12 x 32 GBDDR4-2933MT/s HMA84GR7CJR4N-WM (.....)
      Line break after each ":" in :
      Compiler Notes:
      Disk Notes:
      (...)

      A header before the table with the summary of results, the way it is you are reading the system information and suddenly are presented with a huge table
      You could also break the page, so the summary of results would start in a new page

      The graphs are looking blurry, as they were rendered in the wrong resolution in regards to the page, so they were upscaled

      Comment


      • #4
        Originally posted by andrei_me View Post
        Michael feedback of the PDF results:


        I like one line for sentence/information, so my feedback might be biased

        I would change the part "The results with the greatest spread from best to worst included" into bullet lists or a table
        • Tungsten Renderer (Scene: Volumetric Caustic) at3.53x,
        • MBW (Test: Memory Copy - Array Size: 4096 MiB) at 3.22x,
        • Parboil (Test: OpenMP MRI Gridding) at 3.02x,
        • Hackbench (Count: 32 - Type: Process) at 3.01x,
        • Tachyon (Total Time) at 2.62x
        • C-Ray (Total Time - 4K, 16 Rays PerPixel) at 2.37x
        • dav1d (Video Input: Chimera 1080p) at 2.33 (.....)

        Same for "Test Systems":
        • Processor: 2 x Intel Xeon Platinum 8280 @ 4.00GHz (56 Cores / 112 Threads)
        • Motherboard: GIGABYTEMD61-SC2-00 v01000100 (T15 BIOS)
        • Chipset: Intel Sky Lake-E DMI3 Registers
        • Memory: 12 x 32 GBDDR4-2933MT/s HMA84GR7CJR4N-WM (.....)
        Line break after each ":" in :
        Compiler Notes:
        Disk Notes:
        (...)

        A header before the table with the summary of results, the way it is you are reading the system information and suddenly are presented with a huge table
        You could also break the page, so the summary of results would start in a new page

        The graphs are looking blurry, as they were rendered in the wrong resolution in regards to the page, so they were upscaled
        Thanks. I just uploaded the PDF and added line breaks to the automated summary portion. Line breaks under system information though would take up too much space especially for large systems, so for now that is remaining the same.

        In the updated PDF should also be clearer graph rendering. With 9.4 I added support for using rsvg2 or inkscape for converting SVG graph markup to PNGs for PDFs as an alternative to my own custom-written SVG-to-PNG layer. My layer renders the graphs clearer but I like the text layout handling better on inkscape/rsvg2, but in any case with latest fallback is my layer now so the graphs are cleaner.
        Michael Larabel
        https://www.michaellarabel.com/

        Comment


        • #5
          Originally posted by Michael View Post
          Thanks. I just uploaded the PDF and added line breaks to the automated summary portion..
          What do you think of putting the N times before the test?

          E.g:
          Tungsten Renderer (Scene: Volumetric Caustic) at 3.53x
          MBW (Test: Memory Copy - Array Size: 4096 MiB) at 3.22x
          Parboil (Test: OpenMP MRI Gridding) at 3.02x
          Hackbench (Count: 32 - Type: Process) at 3.01x

          Would become this:
          3.53x on Tungsten Renderer (Scene: Volumetric Caustic)
          3.22x on MBW (Test: Memory Copy - Array Size: 4096 MiB)
          3.02x on Parboil (Test: OpenMP MRI Gridding)
          3.01x on Hackbench (Count: 32 - Type: Process)

          As english is not my native language, "on" might be the wrong preposition

          Also, the first item in the list (Tungsten) is missing the bullet/marker

          And before I forget, awesome feature this PDF export, congrats

          Comment


          • #6
            Originally posted by andrei_me View Post
            The graphs are looking blurry, as they were rendered in the wrong resolution in regards to the page, so they were upscaled
            This
            ## VGA ##
            AMD: X1950XTX, HD3870, HD5870
            Intel: GMA45, HD3000 (Core i5 2500K)

            Comment


            • #7
              AFAICS the normalized percentages of the timed compilations on page 9 in the pdf report are wrong. The slower results should be > 100%.

              Comment


              • #8
                Originally posted by knweiss View Post
                AFAICS the normalized percentages of the timed compilations on page 9 in the pdf report are wrong. The slower results should be > 100%.
                Just the way the normalized value is shown for lower-is-better results with based upon feedback from others wanting '100%' always to be the best rather than inverted for LIB data.
                Michael Larabel
                https://www.michaellarabel.com/

                Comment


                • #9
                  I can see the intel HQ skyscraper in silicon valley from my window, they are currently loading up the blue miltary helicopters to pickup michael for re-education on article posting.

                  Goodluck man.
                  I'll miss your articles.

                  Comment


                  • #10
                    Hi Michael,
                    I hope that you, wife and new baby are doing well. I am a grandfather, who after 60 years in IT software and hardware engineering took retirement. I am very impressed and pleased with your presentations and publications. Reason for my post is the following:

                    I was looking at the PDF header information which described the setup. Did I miss something because I did not note the compiler settings or the host operating system. I am more interested in the compiler settings than the host OS. As a matter of curiosity, were the tests performed using Clear Linux?

                    Regards from Canada

                    Comment

                    Working...
                    X