Announcement

Collapse
No announcement yet.

100+ Benchmarks Of Amazon's Graviton2 64-Core CPU Against AMD's EPYC 7742

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

  • 100+ Benchmarks Of Amazon's Graviton2 64-Core CPU Against AMD's EPYC 7742

    Phoronix: 100+ Benchmarks Of Amazon's Graviton2 64-Core CPU Against AMD's EPYC 7742

    Last week Amazon AWS promoted their Graviton2 instances to general availability status with a variety of different sized EC2 instances as well as a bare metal instance for tapping the full potential of their new SoC that features 64 Arm Neoverse N1 cores. Last week we ran through many benchmarks looking at Graviton2 on EC2 and bare metal performance while here is a follow-up article with more benchmarks and looking at how the sixty-four core Arm Graviton2 compares to AMD's EPYC 7742 64-core CPU with and without SMT.

    http://www.phoronix.com/vr.php?view=29206

  • #2
    If Amazon's CPU offers better perf/W, it's a big win, despite the overall slower performance.

    Comment


    • #3
      It looks like Hyper-threading is useless for HPC.

      Comment


      • #4
        To be honest, it is impressive to see an ARM based product winning against high-end x86-64 competitors in 10% of benchmarks. Assuming that it is using a little less power, and is a little bit cheaper, the perf/W and perf/$ should be competitive or desirable.

        Obviously Zen 3 coming soon is a big leap again on the Epyc side of things, but there will be a Graviton 3 next year as well, I would assume.

        Comment


        • #5
          Originally posted by Setif View Post
          It looks like Hyper-threading is useless for HPC.
          That's expected. HT only offers extra execution units, but shares the front end. When there's intensive memory access, the front end becomes a bottleneck and the execution units starve.
          When the data fits in the cache, HT can match the performance of physical cores though.

          Comment


          • #6
            ...Im sorry but I cant get rid of this thought:
            Future me: *using Amazon Graviton2 for CFD Simulations*
            Amazon Firestick Ad: "Hey you like high reynoldsnumbers? Why dont you get the prime deal for the latest Ansys" .
            ...creepy future ahead.

            Comment


            • #7
              Originally posted by CochainComplex View Post
              ...Im sorry but I cant get rid of this thought:
              Future me: *using Amazon Graviton2 for CFD Simulations*
              Amazon Firestick Ad: "Hey you like high reynoldsnumbers? Why dont you get the prime deal for the latest Ansys" .
              ...creepy future ahead.
              Don't worry about it, Amazon doesn't build these for consumers. They just need something better tailored for their usage patterns at a possible lower cost.

              Also, if you're into CFD simulations, would you rather see a cat litter box commercial instead? :P

              Comment


              • #8
                Michael,
                could you be so kind and provide new tests on your POWER9 machines? IIRC you have 4 core and 2x 18 cores provided by Raptor? If so, then having updated test results on those would be fantastic since comparing old llvm build (6.0) with new llvm build (10.0) is pretty unrealistic. I know power9 results will be a bit of shame and power10 is planned for next year, but still how much low perf it will be, that is the question.

                Yes, I'm developer primary interested in compilation benchmarks.

                Thanks!
                Karel

                Comment


                • #9
                  Originally posted by bug77 View Post
                  If Amazon's CPU offers better perf/W, it's a big win, despite the overall slower performance.
                  Seems like another technology that's been overhyped for years and is just always going to have a corner use case where failing to provide comparable performance to x86 is acceptable.

                  Comment


                  • #10
                    Originally posted by bug77 View Post

                    That's expected. HT only offers extra execution units, but shares the front end. When there's intensive memory access, the front end becomes a bottleneck and the execution units starve.
                    When the data fits in the cache, HT can match the performance of physical cores though.
                    This is not true, SMT or HT does not add execution units. The reason that SMT/HT does not help HPC is because these applications are bottlenecked by the FP execution units or memory bandwidth. Sometimes there is a performance penalty for HPC applications when SMT/HT is enabled, because of cache contention.

                    Comment

                    Working...
                    X