Announcement

Collapse
No announcement yet.

A Look At The Xeon Gold 6138 + Tyan GT24E-B7106 1U Linux Server Performance

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

  • A Look At The Xeon Gold 6138 + Tyan GT24E-B7106 1U Linux Server Performance

    Phoronix: A Look At The Xeon Gold 6138 + Tyan GT24E-B7106 1U Linux Server Performance

    Last week I began testing the Tyan GT24E-B7106, a 1U barebones server designed for Intel's new Xeon Scalable processors. I am still carrying out many benchmarks of the Tyan GT24E-B7106 paired with two of the Xeon Gold 6138 CPUs, but for those curious about the Linux performance potential of this server when slotting in 96GB of DDR4-2666 RDIMMs and these two CPUs that yield a combined total of 40 cores / 80 threads, here are some initial benchmarks.

    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
    Thanks for running these benchmarks and for using Clear Linux, which has much better open-source tuning for newer chips than older versions of Ubuntu.

    Incidentally, if GROMACS is not in the PTS then you might want to add a newer version of it with AVX-512 support for additional testing. It's a popular HPC workload that AMD went out of its way to demonstrate at the Epyc launch. The AVX-512 performance numbers are extremely strong and it would be a good real-world test to show AVX-512 in action.

    Comment


    • #3
      Typo:

      Originally posted by phoronix View Post
      Coming up in the days ahead are still a number of other intesting tests

      Comment


      • #4
        2 questions, if you don't mind:

        1. Do you have to return this little beast?
        2. Have you got the EPYC server yet? it would be very interesting to put those 2 against each other..

        Comment


        • #5
          Huh. Seems kinda fast I guess. I suppose you could say 24.75s compile time is pretty quick.

          Comment


          • #6
            Originally posted by schmidtbag View Post
            Huh. Seems kinda fast I guess. I suppose you could say 24.75s compile time is pretty quick.
            @ 2.5k ish per chip it bloody well should be

            Comment


            • #7
              I was promised LLVMpipe numbers!

              Comment


              • #8
                Originally posted by LinuxID10T View Post
                I was promised LLVMpipe numbers!
                Yes, they are coming. LLVMpipe + SWR in its own article.
                Michael Larabel
                https://www.michaellarabel.com/

                Comment


                • #9
                  Originally posted by Michael View Post

                  Yes, they are coming. LLVMpipe + SWR in its own article.
                  Sounds awesome! I've noticed that both LLVMpipe and SWR seem to use AVX extensively when available (well, SWR exclusively) I think it will be a slaughtering if you end up testing Epyc with them.

                  Comment


                  • #10
                    Originally posted by LinuxID10T View Post

                    Sounds awesome! I've noticed that both LLVMpipe and SWR seem to use AVX extensively when available (well, SWR exclusively) I think it will be a slaughtering if you end up testing Epyc with them.
                    Well seeing anandtech data and some other enterprise sites reviews seems Skylake-X have a wider FPU but AMD Zeppelin FPU have more raw horse power, so far it seems in regular AVX loads dual Epycs piss all over dual Xeons but maybe intel can catch up with optimized binaries using AVX-512 or at least get closer.

                    Btw Michael if you test this compiling yourself make sure Epyc is optimized with GCC and/or Clang not ICC since the few tests lying around show ICC slows down Epyc(not a surprise to anyone LOL) performance but it boost around 25% the xeons. dunno is tricky to find a balance
                    Last edited by jrch2k8; 24 August 2017, 11:23 AM. Reason: Also I believe pushing hard AVX-512 will require a portable fission reactor, seeing a dual Xeons 8176 need around 125-150w under regular load than a dual Epyc, just imaging punching the gas on the AVX

                    Comment

                    Working...
                    X