SMT Performance Benchmarks Continue To Show Benefit With AMD Zen 5/5C

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • andre30correia
    Senior Member
    • May 2015
    • 1154

    #11
    Originally posted by avis View Post

    Apple is a performance leader and they have zero HT/SMT in their CPUs. And Intel is doing exactly the same.

    x86 uArchs trail Apple by a large margin both in terms of raw performance and efficiency.
    where apple have more performance than amd? even a simple ryzen 7 5700u super cheap have better multi performance and gpu performance

    Comment

    • espi
      Phoronix Member
      • Mar 2022
      • 64

      #12
      Originally posted by andre30correia View Post

      where apple have more performance than amd? even a simple ryzen 7 5700u super cheap have better multi performance and gpu performance
      Laptop M3 Max (12P+4E) competes with the desktop Ryzen 9 7900X (12P+SMT), on both 1T an nT. Of course its a stupid comparison, but the iGPU in M3 Max is also faster than the iGPU of Ryzen 9 7900X. Any decent discrete GPU crushes the Apple iGPU of course.

      The GPU in M3 Max also crushes the one in the Ryzen 7 5700U. It's compatibility is very spotty compared to the AMD GPU, but when it works its easily quadruple the performance (I did own a laptop with Ryzen 7 Pro 5850U, then switched to M3 Max).

      The problem with Apple is that their solutions are incredibly inflexible and cost ineffective. For the cost a Macbook with M3 Max you can easily get a 7950X (16P+SMT) that does crush the M3 Max.
      Last edited by espi; 02 August 2024, 02:23 PM.

      Comment

      • drakonas777
        Senior Member
        • Feb 2020
        • 532

        #13
        Originally posted by avis View Post

        x86 uArchs trail Apple by a large margin both in terms of raw performance and efficiency.
        I can let you slide with efficiency (in stand by and light threads mostly), but Apple M series is nowhere near to be ahead by large margin in terms of absolute raw performance unless comparing something that is accelerated by the specialized hardware in M.

        Comment

        • milkylainen
          Senior Member
          • Mar 2012
          • 1106

          #14
          As the frontend and resources get beefier, the benefit of SMT usually increases, not the opposite?
          At the same time, if you can fill every physical resource all the time without another context, then you'd be better without it?

          I mean, sure, the tradeoff calculation needs to be done against the transistor budget,
          but saying that SMT has no place because Intel is doing away with it isn't necessarily the right answer for everyone else.

          Comment

          • schmidtbag
            Senior Member
            • Dec 2010
            • 6618

            #15
            Clearly, AMD feels SMT is worth the transistor budget if even their shrunken-down Zen4c and Zen5c cores still use it. From what I understand, it involves a lot more transistors than HT, so if they were better off ditching it, I think they would by now.

            Comment

            • varikonniemi
              Senior Member
              • Jan 2012
              • 1102

              #16
              Originally posted by fitzie View Post
              no one in risc-v or arm are going anywhere near smt. as jim keller says, there's better things to do with the transistor budget.
              like keeping pipelines unused? That's winning if you have free silicon to deploy your transistors on.

              And unless your separate core can achieve similar idle consumption as two HT cores, then your idle consumption will not matter, and as we see here, at least AMD design is more efficient when using HT
              Last edited by varikonniemi; 02 August 2024, 03:14 PM.

              Comment

              • Errinwright
                Senior Member
                • Aug 2023
                • 189

                #17
                Originally posted by avis View Post

                Apple is a performance leader and they have zero HT/SMT in their CPUs. And Intel is doing exactly the same.

                x86 uArchs trail Apple by a large margin both in terms of raw performance and efficiency.
                Calling Apple the performance leader is simply ridiculous - likely your worst take thus far. There are differences in perf./price stemming from process node differences, and differences in the latter implies significant manufacturing cost. The largest gains observed on Apple are all the hardware level accelerators they implement (which look great on paper, but take up critical die area for specific use-cases). Their battery life comparisons are a joke. Going passive cooling (and embracing perf. throttling) gained them 60-90minutes additional battery duration. Their largest gains comes from the kernel not continually being hassled by thirdparty implementations sending interrupts. A major reason Microsoft is pushing Arm on Windows is not because the ISA is so powerful, but because all the third party kernel level implementations that make proper C-states impossible - in short, they want all of it out of the kernel with the Arm laptops. Perhaps this might backfire down the line when(/if) kernel level anticheat is expelled from the kernel making Windows obsolete for a majority of gamers.

                Comment

                • Jorgp2
                  Junior Member
                  • Jan 2022
                  • 17

                  #18
                  Originally posted by V1tol View Post

                  We are in 2024. Stupid Chrome runs 22 processes with only 2 tabs. Open more tabs, launch some Electron crap - and all your cores will have something to do. And people also use Windows which has tons of background crap... And SMT helps here a lot because those processes don't do much so hardware scheduling on the same physical core doesn't hurt performance.
                  I don't think you understand the difference between multiprocessing and multi threading.

                  Comment

                  • kylew77
                    Senior Member
                    • Jul 2017
                    • 1138

                    #19
                    Symmetric Multithreading, or "Why are only half of my CPUs used?"

                    Some CPUs use Symmetric Multithreading (SMT; Intel's implementation is known as "Hyper-Threading"). In this case, one physical processor presents an extra logical processor to the OS - shown as a separate CPU in dmesg(8) and tools like top(1). These do not have full CPU resources but are there to allow sharing part of a single core's resources with more than one concurrent process.

                    This feature can improve performance for some workloads but reduces it for others.

                    SMT has been involved in various CPU vulnerabilities, in particular relating to speculative execution. This can result in processes learning information about other processes which they should not have access to. To mitigate this, OpenBSD disables running code on detected SMT "virtual" cores by default.
                    Source: https://www.openbsd.org/faq/faq10.html

                    Once again we see a security vs performance trade off, which one you choose is up to you.

                    Comment

                    • Anux
                      Senior Member
                      • Nov 2021
                      • 1960

                      #20
                      Overall the CPU package power consumption averaged out to being the same across both runs.
                      Hm, puzzling. I expected a pronounced difference and also the efficiency should be closer together. Is the CPU unable to put the unused parts to sleep? Or is it clocking higher in noSMT? No frequency stats sadly.

                      Comment

                      Working...
                      X