Announcement

Collapse
No announcement yet.

AMD P-State Preferred Core Patches For Linux Updated, Will Be Enabled By Default

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

  • AMD P-State Preferred Core Patches For Linux Updated, Will Be Enabled By Default

    Phoronix: AMD P-State Preferred Core Patches For Linux Updated, Will Be Enabled By Default

    AMD's Preferred Core feature continues working its way toward the Linux kernel for this functionality that's been around since Zen 2...

    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
    "...if AMD Ryzen processors go in a hybrid design with a mix of normal and "C" energy-efficient cores,..."

    E- and P-cores after Zen 5.

    BTW Back from vacation.
    Last edited by nuetzel; 15 August 2023, 08:45 AM.

    Comment


    • #3
      Too bad that some laptop manufacturers like Lenovo or Asus never enabled ACPI_CPPC on most of their Zen2 laptops, and now refuse to enable it under fallacious arguments like "AMD told us it's better no to" because they obviously don't want to allocate any resource to older devices. Tells us everything we need to know about the future of the RoG AlLy.

      A 4800U is still a fantastic cpu by today standards, but profits matter much more than this world's e-waste problem for this companies.

      Comment


      • #4
        Originally posted by nuetzel View Post
        "...if AMD Ryzen processors go in a hybrid design with a mix of normal and "C" energy-efficient cores,..."

        E- and P-cores after Zen 5.
        That's wrong by phoronix, C-Cores are not more energy efficient, they are more space efficient. You could probably get the same efficency by disabling parts of the L2 cache and restrict the max clock of a normal core.

        Originally posted by citral View Post
        Too bad that some laptop manufacturers like Lenovo or Asus never enabled ACPI_CPPC on most of their Zen2 laptops, and now refuse to enable it under fallacious arguments like "AMD told us it's better no to" because they obviously don't want to allocate any resource to older devices.
        I just recently read somewhere that you can load your own UEFI and activate it. Can't find it now but I leave the searching to you.

        Comment


        • #5
          Originally posted by nuetzel View Post
          "...if AMD Ryzen processors go in a hybrid design with a mix of normal and "C" energy-efficient cores,..."

          E- and P-cores after Zen 5.

          BTW Back from vacation.
          Wouldn't this kill x86-64-v4/AVX512 microarchitecture support?

          Comment


          • #6
            Absolutely hate the concept of P-cores and E-cores which is why I went with AMD in the first place. Hope this doesn't change in the future..

            Comment


            • #7
              Originally posted by vextium View Post

              Wouldn't this kill x86-64-v4/AVX512 microarchitecture support?
              AMD "C" cores are not like Intel "E" cores. They have the same hardware features as the normal Zen cores, but are rearranged in a way that reduces spaces. This has the problem of also limiting frequencies because of thermals, but for an efficient core that is a feature .

              As others has pointed out, AMD also puts less cache on this as another space saving measure.

              Comment


              • #8
                I am running the patch right now, and it seems to be working although does have a couple quirks. I do think this feature will be very useful for those that have CPUs with widely ranging frequencies, where it is a difference of 300MHz+ from best to the median. For others with more balanced CPUs the benefit will be less.

                From my benchmarking core 0 is my best core when it is isolated, however when it isn't it gets saddled with kernel threads and is no longer the most performant core. If I remember correctly when I was benchmarking to test the amd-pstate modes my top 6 cores were 4,6,0 (not isolated),3,2,8. Preferred core does seem to schedule tasks of increasing thread count in roughly that order, although it does seem to like core 8 more than 2 and 3.

                One quirk is that while I was doing a single thread zstd benchmark it was getting moved around between core 4 and 6 rapidly, to the point that tracking core utilization showed a fluctuating 40-60% on either of the cores. This persisted for only one more run, and I haven't been able to get it to repeat.

                ​​​​​​The other quirk is that there is no way to turn it off once booted, that would've been nice to make testing a bit easier.
                ​​

                Comment


                • #9
                  Originally posted by Namelesswonder View Post
                  From my benchmarking core 0 is my best core when it is isolated​​
                  Yes that's what the article refers to with
                  "static core ranking".

                  ​​
                  ​​​​​​The other quirk is that there is no way to turn it off once booted, that would've been nice to make testing a bit easier.​​
                  Are you able to influence it's behavior with
                  Code:
                  powerprofilesctl set power-saver
                  (and the other profiles)

                  Comment


                  • #10
                    Originally posted by jonix View Post

                    AMD "C" cores are not like Intel "E" cores. They have the same hardware features as the normal Zen cores, but are rearranged in a way that reduces spaces. This has the problem of also limiting frequencies because of thermals, but for an efficient core that is a feature .

                    As others has pointed out, AMD also puts less cache on this as another space saving measure.
                    All of that is correct. Also part of shrinking the cache in the Zen4c core design is elimintation of through-silicon-vias (TSVs) for attaching 3D vcache. They are basically pillars of copper dotted throughout the cache to connect the SRAM fabrics of the stacked die. They are relatively massive compared to the nanometer scale structures on the rest of the chip.

                    Comment

                    Working...
                    X