Announcement

Collapse
No announcement yet.

ThunderX2 Getting Big MEMMOVE Performance Boost With Glibc 2.30

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

  • ThunderX2 Getting Big MEMMOVE Performance Boost With Glibc 2.30

    Phoronix: ThunderX2 Getting Big MEMMOVE Performance Boost With Glibc 2.30

    For those running Arm servers powered by the Cavium ThunderX2, a big performance optimization is on the way for Glibc 2.30...

    http://www.phoronix.com/scan.php?pag...ove-glibc-2.30

  • #2
    Umm. Why not make it AArch64-generic? I understand these are custom AArch64 SoCs but they do adhere to the ARM v8 ISA do they not?
    Hate to be the negative guy but the reindent + code rework commit in same is just fugly and lazy.
    Last edited by milkylainen; 05-05-2019, 01:59 PM.

    Comment


    • #3
      Because generic doesn't pay the bills. I'm not being snarky. Just truth. There is a metric crap ton of money to be made in the server world and the HPC world. ANY...ANY advantage or performance tuning that can be had for a specific ARM based SoC such as Cavium is ok by me if it helps them to be more viable against Swindle-tel....err...Intel. So what if the code is SoC specific. ANYONE can look at the damn code and tune it for the generic AArch64 anytime they wish.

      Comment


      • #4
        Originally posted by Jumbotron View Post
        Because generic doesn't pay the bills. I'm not being snarky. Just truth. There is a metric crap ton of money to be made in the server world and the HPC world. ANY...ANY advantage or performance tuning that can be had for a specific ARM based SoC such as Cavium is ok by me if it helps them to be more viable against Swindle-tel....err...Intel. So what if the code is SoC specific. ANYONE can look at the damn code and tune it for the generic AArch64 anytime they wish.
        What you say is tangential. Applying optimizations to AArch64-generic or ThunderX2 target would be the same for "fighting Intel", actually better the former as now more Arch servers can have better performance.

        The only two reasons someone would want to make a target-specific optimization is
        -because the hardware actually differs from the generic
        -because he is an ass

        In this case it's entirely possible that the hardware does genuinely differ enough (cache sizes or some kind of additional hardware accelerator for something)

        Comment


        • #5
          There is still nothing between 4-6GB RAM boards and rack servers? This one definitely have interesting price (3000 USD) especially because it's two socket server with 64 7nm A76-based cores per CPU, but something cheaper and smaller is still needed. And, no, Developerbox with 24 A53 is not looking good at it's price (I also can order Huawei server from local reseller in Russia, but can't order Developerbox).

          Comment


          • #6
            Originally posted by RussianNeuroMancer View Post
            There is still nothing between 4-6GB RAM boards and rack servers? This one definitely have interesting price (3000 USD) especially because it's two socket server with 64 7nm A76-based cores per CPU, but something cheaper and smaller is still needed. And, no, Developerbox with 24 A53 is not looking good at it's price (I also can order Huawei server from local reseller in Russia, but can't order Developerbox).
            Are you sure it's using ARM processors? the images show an intel sticker under the power button. Is Huawei being overly sneaky here, selling ARM servers that look like Intel servers?

            Comment


            • #7
              Cavium Thunder X2 is definetly ARM, not Intel..

              B.t.w. ARM N1 / E1 is comming, it will offer 4-6 GB RAM boards for computing at home

              Comment


              • #8
                Originally posted by milkylainen View Post
                Umm. Why not make it AArch64-generic? I understand these are custom AArch64 SoCs but they do adhere to the ARM v8 ISA do they not?
                Hate to be the negative guy but the reindent + code rework commit in same is just fugly and lazy.
                the ISA might be generic - but the rest of the chip may not be. There could be all sorts of constraints on how memory is moved around (slow/weak/cheap interconnects).

                Comment


                • #9
                  Originally posted by starshipeleven View Post

                  Are you sure it's using ARM processors? the images show an intel sticker under the power button. Is Huawei being overly sneaky here, selling ARM servers that look like Intel servers?
                  Just wrong photo, check specs here: https://e.huawei.com/ru/products/clo...aishan-2280-v2

                  Comment


                  • #10
                    Originally posted by RussianNeuroMancer View Post
                    There is still nothing between 4-6GB RAM boards and rack servers? This one definitely have interesting price (3000 USD) especially because it's two socket server with 64 7nm A76-based cores per CPU, but something cheaper and smaller is still needed. And, no, Developerbox with 24 A53 is not looking good at it's price (I also can order Huawei server from local reseller in Russia, but can't order Developerbox).
                    Can you order from UK? If so, then have a look at avantek: https://store.avantek.co.uk/arm-desktops.html

                    Comment

                    Working...
                    X