Announcement

Collapse
No announcement yet.

AMD Ryzen 9 9950X DDR5-6000 / DDR5-6400 / DDR5-8000 Memory Performance

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

  • AMD Ryzen 9 9950X DDR5-6000 / DDR5-6400 / DDR5-8000 Memory Performance

    Phoronix: AMD Ryzen 9 9950X DDR5-6000 / DDR5-6400 / DDR5-8000 Memory Performance

    For those planning on upgrading to an AMD Ryzen 9 9900 series Zen 5 desktop for Linux use, here are some benchmarks I recently carried out of several different DDR5 memory kits with the AMD Ryzen 9 9950X while looking at the Linux performance under a variety of different workloads.

    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
    So a second faster in kernel compilation and one or two FPS in games?

    Think i'll stick with the max in-spec in-warranty 5200MT/s in my workstation.

    Until AMD decides that EXPO is okay, not risking it.

    Comment


    • #3
      Now, this is my personal experience.
      I hope ddr5 on 9000-series works out better than on the 7000-series.
      My 7950x was a total faff with the memory.

      And AM4 was way, way easier.

      Comment


      • #4
        In all computers, but especially workstations, stability is way more important than raw performance.

        I am looking forward to the day when computers don't have ram modules, just like the Xeon Max has that onboard 64gb ram, MD should do the same to their processors.

        Side note, in a high end server, where there is large amount of ram, in some cases greater than a terabyte, ram accounts for the majority of power consumption,

        Another reason to do away with ram modules.

        Comment


        • #5
          For the RAM comparison, it'd be good to show the "stream" software benchmark compiled with and without -fopenmp.

          Comment


          • #6
            Originally posted by emansom View Post
            i'll stick with the max in-spec in-warranty 5200MT/s in my workstation.
            That's a smart move for sure. Don't forget that for Zen5 the now in-spec memory speed is JEDEC 5600. As a side note, for those running the G.Skill Trident Z Neo, you can tweak your memory timings by following this tutorial from buildzoid, your mileage may vary of course.

            With DDR5-6400 said to be the "sweet spot" for the Zen 5 Ryzen 9000 series...
            Hey Michael, (Sorry I couldn't @ you), was this recently changed by AMD?

            I recall from Level1Techs and Hardware Unboxed that the "sweet spot" for Zen5 was still 6000 MHz memory (According to the reviewers guide)?

            Source; https://youtu.be/IeBruhhigPI?t=1446

            Comment


            • #7
              Originally posted by sophisticles View Post
              In all computers, but especially workstations, stability is way more important than raw performance.

              I am looking forward to the day when computers don't have ram modules, just like the Xeon Max has that onboard 64gb ram, MD should do the same to their processors.

              Side note, in a high end server, where there is large amount of ram, in some cases greater than a terabyte, ram accounts for the majority of power consumption,

              Another reason to do away with ram modules.
              For small pc's that are barely sweating anyway it might be OK, but how would you fit all the required TB's of RAM on a high core count CPU? And how much would that decrease power consumption, since the most of it used up by refreshing those stupid DRAM cells?

              Comment


              • #8
                Originally posted by sophisticles View Post
                In all computers, but especially workstations, stability is way more important than raw performance.

                I am looking forward to the day when computers don't have ram modules, just like the Xeon Max has that onboard 64gb ram, MD should do the same to their processors.

                Side note, in a high end server, where there is large amount of ram, in some cases greater than a terabyte, ram accounts for the majority of power consumption,

                Another reason to do away with ram modules.
                For what it is worth, I have been preaching for a while now that HBM is the future. The DDR technology is merely going through its last iterations before all investments into it have paid out. We will see DDR6, and possibly DDR7, more memory channels and a few new tricks in memory management to go with these last iterations. However, the returns are diminishing and the end is in sight. With ever increasing latencies, and L3 caches being implemented as separate dies now, is HBM the next logical step. It may not be HBM afterall, but a modified variation of it, however with the advances made with HBM, availability of design tools, and experience accumulating around it is it the top candidate for a successor, But if I am wrong and DDR sees many more iterations then computers' main memory is going to get really boring ...

                Comment


                • #9
                  Interesting to see the difference in performance amongst these memory modules, but there are too many factors at play to really get a good idea of what's going on. The difference between 1:1 and 1:2 modes is one factor. The memory transaction speed is another. The absolute latency is the third.

                  So, the two 8000 modules had higher transaction speed than the 6000 and 6400, but had lower absolute latency--though they had to operate at 1:2 mode.
                  The 6000 and the 6400 had the same absolute latency (but slower than the 8000s), slower transaction speed, but operated in the faster 1:1 mode.

                  Maybe try running the two 8000 modules at CL40 to remove the absolute latency mismatch? Then you'll be seeing a more fair comparison between transaction speed and 1:2/1:1 ratio. (it's still two variables changing, but it's not as bad).

                  Obviously, that's not how you'd want to run a final system to get best performance, but it would be useful as a *benchmark* to be able to understand the relative importance of the different characteristics of the memory system.

                  Comment


                  • #10
                    I actually have a brand new MSI MAG X670E Tomahawk WiFi AM5 motherboard in my closet, but it's very difficult to justify replacing my ASUS TUF Gaming X570/Ryzen 7 5700X workstation as I'd still have to spend another $500 or more for a new CPU and DDR5, and perhaps cooler.

                    While the AM5 system would of course be faster, this is one of those cases where AMD did such a great job with their previous generation platform that upgrading is not a straightforward decision. So I keep going back and forth when I read articles like this - enticed by the excitement of a new and better build, but never quite able to overcome the extra expense barrier, and the fact that the remnants of my perfectly functional AM4 system would be relegated to permanent storage.

                    By the way I only have the motherboard because, after much consultation, a friend decided they wanted me to build them a powerful desktop system instead of buying another laptop. But then after they bought the motherboard they changed their mind and decided on the laptop after all and it was too late to return the MB. So I bought it from them because, well, that's just the kind of guy I am and couldn't bear the fact that they'd spent so much money on something they could never use.

                    Comment

                    Working...
                    X