CachyOS Update Now Uses AutoFDO-Optimized Kernel, Rusticl Driver

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • phoronix
    Administrator
    • Jan 2007
    • 67332

    CachyOS Update Now Uses AutoFDO-Optimized Kernel, Rusticl Driver

    Phoronix: CachyOS Update Now Uses AutoFDO-Optimized Kernel, Rusticl Driver

    The CachyOS December 2024 update is out today as the newest monthly release to this performance-optimized, Arch Linux based operating system...

    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
  • Errinwright
    Senior Member
    • Aug 2023
    • 188

    #2
    God damn I love CachyOS

    Comment

    • Setif
      Senior Member
      • Feb 2016
      • 302

      #3
      I hope to see a comparison between CachyOS and Clear Linux.

      Comment

      • Michael
        Phoronix
        • Jun 2006
        • 14308

        #4
        Originally posted by Setif View Post
        I hope to see a comparison between CachyOS and Clear Linux.
        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
        Michael Larabel
        https://www.michaellarabel.com/

        Comment

        • isaacx123
          Junior Member
          • Nov 2020
          • 29

          #5
          I am kind of confused, according to their blog post AutoFDO is supported on Zen3 Epyc, Zen4 and Zen5, so it isn't supported on Zen3 consumer SKUs? Like the 5800X3D for example?

          Comment

          • leandrolnh
            Junior Member
            • May 2015
            • 6

            #6
            Originally posted by isaacx123 View Post
            I am kind of confused, according to their blog post AutoFDO is supported on Zen3 Epyc, Zen4 and Zen5, so it isn't supported on Zen3 consumer SKUs? Like the 5800X3D for example?
            Probably because AutoFDO needs on AMD the Branch Sampling (BRS) feature, likely available on Zen 3 only on Epyc.

            Comment

            • CommunityMember
              Senior Member
              • Oct 2019
              • 1375

              #7
              So the optimization is based on running the benchmarks that some sites use in order to make the benchmark results look better, rather than show real world usage. Got it.

              In the 80's, a certain company optimized their compiler for a certain Fortran benchmark. And in 2000, a certain company optimized their database for a certain database benchmark workload. What is old is new again (possibly attributed to Twain: "lies, damn lies, and statistics").

              It should be noted that in both of those cases, once the issue of optimizing to the benchmark was identified, those performing the evaluation threw the results out as being misleading. We shall see if Phoronix has the same approach.
              Last edited by CommunityMember; 21 December 2024, 03:09 PM.

              Comment

              • ptr1337
                Senior Member
                • Dec 2021
                • 216

                #8
                Originally posted by isaacx123 View Post
                I am kind of confused, according to their blog post AutoFDO is supported on Zen3 Epyc, Zen4 and Zen5, so it isn't supported on Zen3 consumer SKUs? Like the 5800X3D for example?
                That is only if you want to generate an own profile. Zen3 does not support branch sampling (the consumer CPUs, Epyc Zen3 should support it)

                Comment

                • ptr1337
                  Senior Member
                  • Dec 2021
                  • 216

                  #9
                  Originally posted by CommunityMember View Post
                  So the optimization is based on running the benchmarks that some sites use in order to make the benchmark results look better, rather than show real world usage. Got it.

                  In the 80's, a certain company optimized their compiler for a certain Fortran benchmark. And in 2000, a certain company optimized their database for a certain database benchmark workload. What is old is new again (possibly attributed to Twain: "lies, damn lies, and statistics").

                  It should be noted that in both of those cases, once the issue of optimizing to the benchmark was identified, those performing the evaluation threw the results out as being misleading. We shall see if Phoronix has the same approach.
                  Why?
                  We have just generated a little script, which runs the profiling for around 20 minutes with various different workloads. The "cachyos-benchmarker" contains a bunch of real world workloads (compilation, compression, blender, x265).
                  Additionally in this script we are running a bunch of sysbench workloads to have syscalls and others, rg, find.

                  How you would generate a profile for PGO/AutoFDO? Please let me know, if you have a better workload.

                  Comment

                  • AlDunsmuir
                    Junior Member
                    • Oct 2013
                    • 42

                    #10
                    Seems more like NASCAR's “If you ain't cheatin' you ain't tryin'”.

                    If benchmarks are not a representative workload of real usage, what purpose DO they serve?

                    Comment

                    Working...
                    X