Announcement

Collapse
No announcement yet.

Ubuntu 18.04 Versus Six Other Linux Distributions On AMD EPYC

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

  • Ubuntu 18.04 Versus Six Other Linux Distributions On AMD EPYC

    Phoronix: Ubuntu 18.04 Versus Six Other Linux Distributions On AMD EPYC

    With Ubuntu 18.04 LTS set to be released next month and its final package configuration quickly falling into place, we have begun firing up some benchmarks for seeing how this Ubuntu 18.04 "Bionic Beaver" release is comparing to various other Linux distributions. Up first as part of this series of benchmarks is using an AMD EPYC workstation/server for seeing how the Ubuntu 18.04 LTS performance compares to six other Linux distributions.

    http://www.phoronix.com/vr.php?view=26113

  • #2
    Was this done on the desktop or server version of ubuntu? Those SQL tests are interesting, since I'm planning a server upgrade both with software and hardware. According to this I would get less performance than with 16.04?
    Desktop Environment poll:
    https://www.phoronix.com/forums/foru...de-do-you-like

    Comment


    • #3
      Typos:

      Originally posted by phoronix View Post
      kernel and GCC .7.3.
      Originally posted by phoronix View Post
      ended up causing this SQL databse system to run the fastest on it.

      Comment


      • #4
        Of course Intel's Clear Linux is going to win: It's the only one using the performance governor by default!

        BTW, if anyone wants to get better performance AND latency out of their Intel CPUs, I'd recommend using the performance governor while disabling Intel's P-State driver; like this:

        Code:
        intel_pstate=disable
        Even though Intel also calls it performance, their driver still tries to dynamically scale the CPU frequency, which manifests itself in increased latency response.

        Comment


        • #5
          Originally posted by Linuxxx View Post
          Of course Intel's Clear Linux is going to win: It's the only one using the performance governor by default!

          BTW, if anyone wants to get better performance AND latency out of their Intel CPUs, I'd recommend using the performance governor while disabling Intel's P-State driver; like this:

          Code:
          intel_pstate=disable
          Even though Intel also calls it performance, their driver still tries to dynamically scale the CPU frequency, which manifests itself in increased latency response.
          Intel CPUs have a knob which allows to bias them towards either power saving or more performance.
          you need to build the "x86_energy_perf_policy" tool available in the linux kernel tree and run it with parameters "--cpu 0-7 --all performance" to get most perf out of the system (I believe the clear linux kernel does that by default, while the vanilla upstream kernel sets it to "balanced")

          Comment


          • #6
            "When it came to losses, Manjaro 17.1.6 was losing most often with finishing in last 10 times and then Debian Testing with seven finishes in last and Debian Testing with six losses."

            Debian Testing did so bad it was on the losers list twice.

            Comment


            • #7
              Originally posted by Linuxxx View Post
              Of course Intel's Clear Linux is going to win: It's the only one using the performance governor by default!

              BTW, if anyone wants to get better performance AND latency out of their Intel CPUs, I'd recommend using the performance governor while disabling Intel's P-State driver; like this:

              Code:
              intel_pstate=disable
              Even though Intel also calls it performance, their driver still tries to dynamically scale the CPU frequency, which manifests itself in increased latency response.
              We turn off the power management all the way across the board to maintain consistent latency. BIOS, OS, etc. Agree with the above remarks.

              Comment


              • #8
                Originally posted by Linuxxx View Post

                Even though Intel also calls it performance, their driver still tries to dynamically scale the CPU frequency, which manifests itself in increased latency response.
                That was actually easy to notice in measuring low level execution times. And it surprised me, since I thought in performance mode, the CPU would regulate frequency in hardware automatically, and not require lengthy interrupts. Which made me wonder if that’s avoidable and could be considered a performance bug, as it not only affects latency, but also average performance (the sum of interrupt times appeared to be relevant).

                Comment


                • #9
                  Originally posted by NateHubbard View Post
                  "When it came to losses, Manjaro 17.1.6 was losing most often with finishing in last 10 times and then Debian Testing with seven finishes in last and Debian Testing with six losses."

                  Debian Testing did so bad it was on the losers list twice.
                  Ha, ha, i even counted 11 loses for Manjaro... whole sentence sounds like "too much beers"

                  By looking at that overview i would say how all distros loses somewhere, 4 to 11 times loses any distro have. Also i see, how only Debian and Ubuntu run all tests and all else distros failed to run couple for some reason
                  Last edited by dungeon; 03-15-2018, 02:23 PM.

                  Comment


                  • #10
                    interesting that Manjaro was often either last or second only to Clear Linux.
                    I wonder exactly what the devs are doing/aiming for.

                    Comment

                    Working...
                    X