Announcement

Collapse
No announcement yet.

A Look At Where The P-State Linux Driver Does Bad Against CPUFreq, Clear Linux Tests

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

  • A Look At Where The P-State Linux Driver Does Bad Against CPUFreq, Clear Linux Tests

    Phoronix: A Look At Where The P-State Linux Driver Does Bad Against CPUFreq, Clear Linux Tests

    I'm still running more benchmarks in investigating the Core i5 7600K Linux performance and with even its graphics performance being slower than Skylake. I fired up Clear Linux on this Kaby Lake system this weekend and it's indeed faster than Ubuntu, though there still is some sort of fundamental issue at play with these new CPUs on Linux. But what is clear is that there are cases where the P-State CPU frequency scaling driver does perform very poorly over the mature, generic CPUFreq scaling driver...

    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
    Do AMD CPUs also benefit from Clear Linux's optimization?

    Comment


    • #3
      Does CPUFreq enable turboboost or is it just P-state? If I go to performance for P-state on my i5-6300HQ then it seems to keep the CPU's constantly between max-freq and max-turbo boost... If I go to CPUFreq they constantly settle on max-freq. How on earth can P-state possibly be slower if it utilizes turbo-boost?

      Comment


      • #4
        Originally posted by milkylainen View Post
        Does CPUFreq enable turboboost or is it just P-state? If I go to performance for P-state on my i5-6300HQ then it seems to keep the CPU's constantly between max-freq and max-turbo boost... If I go to CPUFreq they constantly settle on max-freq. How on earth can P-state possibly be slower if it utilizes turbo-boost?
        one thing to remember is that acpi cpufreq lies to you. the only way to really look at actual achieved frequency (which is a backwards looking thing, and pretty much a time average) is the turbostat utility (or powertop).. and yes turbo boost is used with either.
        Last edited by arjan_intel; 15 January 2017, 12:09 PM.

        Comment


        • #5
          Originally posted by arjan_intel View Post

          one thing to remember is that acpi cpufreq lies to you. the only way to really look at actual achieved frequency (which is a backwards looking thing, and pretty much a time average) is the turbostat utility (or powertop).. and yes turbo boost is used with either.
          Mkay. So what you are saying is that acpi cpufreq can't tell if the cpu is in a turbo boosted frequency or not? Ergo, although acpi cpufreq looks like it "tops out" at max-freq, turbo boosted frequencies are still used? I'm sorry. I find all the scaling issues really confusing sometimes.

          Comment


          • #6
            Originally posted by milkylainen View Post

            Mkay. So what you are saying is that acpi cpufreq can't tell if the cpu is in a turbo boosted frequency or not? Ergo, although acpi cpufreq looks like it "tops out" at max-freq, turbo boosted frequencies are still used? I'm sorry. I find all the scaling issues really confusing sometimes.
            more or less; to acpi cpufreq "turbo" appears as "maximum + 1 Mhz".
            CPU frequency is a complex beast (the OS can ask for things, but what you get is quite often different based on a set of rules, like often all cores get the maximum of what each individual core requested)

            best to look at turbostat output.. that measures actuals.

            Comment


            • #7
              Hey man, remember to disable Speed Shift 2.0 Technology in the BIOS if you want achieve the classical behaviour (C-states + EIST). Maybe the issue lies here.
              Last edited by lowenz; 15 January 2017, 02:08 PM.

              Comment


              • #8
                Originally posted by wdb974 View Post
                Do AMD CPUs also benefit from Clear Linux's optimization?
                I'm not even sure their kernel is built with AMD hardware support, haven't checked recently though.
                Michael Larabel
                https://www.michaellarabel.com/

                Comment


                • #9
                  Originally posted by wdb974 View Post
                  Do AMD CPUs also benefit from Clear Linux's optimization?
                  it should, but as michael said it maybe isnty even supported

                  Comment


                  • #10
                    Originally posted by Michael View Post

                    I'm not even sure their kernel is built with AMD hardware support, haven't checked recently though.
                    I see. At any rate, it's nice to see optimization that are actually reflected in benchmarks.

                    Comment

                    Working...
                    X