Announcement

Collapse
No announcement yet.

Clear Linux Can Run On AMD's EPYC Platform With Competitive Performance

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

  • Clear Linux Can Run On AMD's EPYC Platform With Competitive Performance

    Phoronix: Clear Linux Can Run On AMD's EPYC Platform With Competitive Performance

    As part of our ongoing AMD EPYC Linux benchmarking, I've been working this week on a cross-distribution GNU/Linux comparison followed by some BSD testing... Of course, I couldn't help but to see if Intel's performance-oriented Clear Linux distribution would run on the AMD EPYC server...

    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
    Rather than AMD attempt to produce an "optimized" Linux distro themselves, I would much prefer they partner with some of the base builders like Debian, RedHat, SUSE or Arch and come up with some common enhancements that work for all users of AMD CPU hardware.

    Comment


    • #3
      IIRC I remember reading somewhere that one of the Zen design goals was to deliver excellent performance using intel Haswell optimized (i.e. existing) binaries. They realized that much of the current software ecosystem has been compiled with intel optimizations, and therefore it's not realistic to demand Zen-optimized binaries on launch day, because they simply wouldn't exist yet for the reviewers doing benchmarks, nor for the real world datacenter customers.

      Comment


      • #4
        Please also test NUMA Interleave All for both AMD and Intel. Please test timed linux compilation. Thanks.
        ## VGA ##
        AMD: X1950XTX, HD3870, HD5870
        Intel: GMA45, HD3000 (Core i5 2500K)

        Comment


        • #5
          It would be nice to see some benchmarks that really investigate the AVX2/FMA3 performance. I suspect that a lot of the benchmark results that show intel parts winning by a large margin are caused by the AMD cpu's not utilizing AVX2 properly (for that application).

          Comment


          • #6
            Originally posted by lucasbekker View Post
            It would be nice to see some benchmarks that really investigate the AVX2/FMA3 performance. I suspect that a lot of the benchmark results that show intel parts winning by a large margin are caused by the AMD cpu's not utilizing AVX2 properly (for that application).
            Zen cores differ from Intel cores significantly in this. While they both support AVX2/FMA the internal implementation is distinct.

            AMD decided to implement 256-bit operations using two 128-bit wide execution units.

            For Intel's Skylake-X there's two 256-bit units that can be used together to process AVX512 for processors with less than 10 cores. For 10 and up there's a second AVX512 unit for each core.

            There's a nice dive article about this: https://www.anandtech.com/show/11550...7800x-tested/3

            Exploiting those possibilities in the high end Skylake-x (and new Xeon) cores will require changes to software. Amusingly Intel doesn't even provide a hardware CPUID facility to detect the second AVX512 core https://twitter.com/stephentyrone/st...45421820080129

            Comment


            • #7
              This may be an ultimate test of commitment of the Clear Linux guys, pre- and post-Spectre, on Intel and AMD.

              Comment

              Working...
              X