Announcement

Collapse
No announcement yet.

AMD Rembrandt CPUFreq vs. AMD P-State Linux Testing

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

  • AMD Rembrandt CPUFreq vs. AMD P-State Linux Testing

    Phoronix: AMD Rembrandt CPUFreq vs. AMD P-State Linux Testing

    Before getting busy with the AMD Ryzen 7000 "Zen 4" desktop testing, I recently wrapped up some benchmarks looking at the ACPI CPUFreq vs. AMD P-State frequency scaling drivers and various governor options for the AMD Ryzen 6000 "Rembrandt" mobile SoCs. If you are curious about the impact of CPUFreq/P-State and the various governors for the latest AMD laptops running Linux, this round of testing is for you.

    https://www.phoronix.com/review/amd-rembrandt-pstate

  • #2
    Seems like a good tradeoff. These things are stupid fast, 10% less fast for 20% or more power saving is a deal I'd take.

    Comment


    • #3
      Originally posted by microcode View Post
      Seems like a good tradeoff. These things are stupid fast, 10% less fast for 20% or more power saving is a deal I'd take.
      Eh, but task energy is actually *higher* for pstate schedutil in many benchmarks.


      If idle power consumption is OK, it seems like pstate-performance is a saner default?

      Comment


      • #4
        Thanks, these tests are very much appreciated. I got the feeling that pstate-performance lost only in fully multi threaded workloads, anywhere else it got the best performance and efficiency even got lower temps and wattage while still having the highest average frequency.
        There should be a mode to auto switch to shedutil on all core load and use performance for everything else.

        Comment


        • #5
          Thanks for the benchmarks. BTW, amd-pstate works with ondemand as well.
          Michael ,
          Did you use a separate kernel build for acpi-cpufreq (CONFIG_X86_AMD_PSTATE=n)? I'm yet to find an easy way to disable amd_pstate (something similar to "intel_pstate=" boot flag).

          Comment


          • #6
            Actually a benchmark with:
            amd-pstate (ondemand/schedutil/performance)
            amd-pstate-epp (powersave with the 5 different preferences power,balance_performance, balance_power, performance, default)
            acpi-cpufreq (schedutil/performance)

            I run the amd-pstate-epp driver since it got released, and it does run very well compared to the amd-pstate.
            cpu gov: powersave
            preference: power

            Comment


            • #7
              Typo:

              Originally posted by phoronix View Post
              The amd-pstate schedutil was sticking to much lower clocks than even acpi-cpufreq schedutil or acpi-cpufreq onemand.

              Comment


              • #8
                Originally posted by ptr1337 View Post
                Actually a benchmark with:
                amd-pstate (ondemand/schedutil/performance)
                amd-pstate-epp (powersave with the 5 different preferences power,balance_performance, balance_power, performance, default)
                acpi-cpufreq (schedutil/performance)

                I run the amd-pstate-epp driver since it got released, and it does run very well compared to the amd-pstate.
                cpu gov: powersave
                preference: power
                How do you enable amd-pstate-epp and set preference?

                Comment


                • #9
                  Originally posted by wooque View Post

                  How do you enable amd-pstate-epp and set preference?
                  Compiling the kernel with the patch, then the amd_pstate_epp driver gets automatically loaded, works equal to the default amd-pstate driver.
                  So at most people simply: amd_pstate.shared_mem=1


                  ​To set a preference you need to use the powersave gov:
                  sudo cpupower -g powersave

                  And then you can set the different preferences with:
                  ```
                  echo power | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
                  ​```

                  Available are:
                  default performance balance_performance balance_power power

                  You can find a working patch here:
                  https://github.com/CachyOS/kernel-pa...ancement.patch

                  Actually some more patches needed for it, to work and patch correctly

                  Comment


                  • #10
                    @ptr1337​ thanks, I thought it's already in kernel, this is too much work for me Gonna wait for it to be merged.
                    Did you noticed any difference compared to acpi-cpufreq powersave?

                    Comment

                    Working...
                    X