Announcement

Collapse
No announcement yet.

NVIDIA GH200 CPU Performance Benchmarks Against EPYC Zen 4 & Xeon Emerald Rapids

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

  • #51
    Originally posted by coder View Post
    I get all that. And, for hybrid workloads that span the CPU and GPU, that stuff is great. The key question is whether the HBM address space shows up in the normal heap for CPU jobs, and I really doubt it does.
    My understanding ist that it does.

    Comment


    • #52
      ARM is growing in the server space with Graviton, Ampere, now Nvidia, and also Google soon. As it is common knowledge now, ARM-based processors has power advantage in the data centers. I have not seen in the HPC space though where performance is equally important as power (maybe more). I know ARM has ambitions for HPC space as well but anybody knows HPC systems using ARM CPUs yet?

      Comment


      • #53
        Originally posted by mrg666 View Post
        ARM is growing in the server space with Graviton, Ampere, now Nvidia, and also Google soon. As it is common knowledge now, ARM-based processors has power advantage in the data centers. I have not seen in the HPC space though where performance is equally important as power (maybe more). I know ARM has ambitions for HPC space as well but anybody knows HPC systems using ARM CPUs yet?
        you find the answer in the link you provided ;-)

        Comment


        • #54
          Originally posted by GPTshop.ai View Post

          you find the answer in the link you provided ;-)
          I should fully wake up before posting next time, thanks! Actually, my dog was nagging me ...

          st,small,507x507-pad,600x600,f8f8f8.u5.jpg

          Comment


          • #55
            Originally posted by mrg666 View Post
            I know ARM has ambitions for HPC space as well but anybody knows HPC systems using ARM CPUs yet?
            Fujitsu's A64FX was the first-ever implementation of ARM's Scalable Vector Extensions and powered Japan's Fukagu supercomputer to the top of the Green500 rankings.

            There is a constant push and pull between budget and architecture in supercomputing, and the passing of time has not made anyone’s arms tired as yet on



            As SVE2 is now a mandatory part of the baseline ARMv9-A specification, it's included in the Neoverse V2 cores used by Nvidia's Grace.

            While searching for a reference to that, I stumbled across this bit of analysis:

            We are still plowing through the many, many presentions from the Hot Interconnects, Hot Chips, Google Cloud Next, and Meta Networking @ Scale conferences

            Last edited by coder; 10 February 2024, 11:55 AM.

            Comment


            • #56
              Well that's even more depressing as I covet this HW.
              Had I bought a hundred or two shares back in the 1990s maybe I'd actually be able to afford
              one of their products now.

              Originally posted by sophisticles View Post

              The system is clearly not meant to be a general purpose computer, as evidenced by the 40+ thousand dollar price tag.

              BTW, there are analysts predicting that NVIDIA's stock could hit over $1200 a share.

              Comment


              • #57
                Originally posted by sophisticles View Post
                So the person that cheer-leads for higher core count Threadrippers now wants to use fewer cores?
                this cpu has 72 cores and is slower than a AMD 64core epyc/threadripper and a threadripper of course has even faster singlecore speed.
                fewer cores is always nice but only if the performance is higher. this 72core gets a bonus for not having hyperthreating.
                but of course you can disable hyperthreating on the threadripper so.
                Phantom circuit Sequence Reducer Dyslexia

                Comment


                • #58
                  Originally posted by GPTshop.ai View Post

                  Comparison of GH200 to alternative systems with the same amount of memory:
                  • Compared to 4x AMD Mi300X, GH200 costs 4x less, consumes 4x less energy and is not far off in terms of performance.
                  • Compared to 5x AMD Mi300A, GH200 costs 3x less, consumes 3x less energy and has at least the same performance.
                  • Compared to 8x RadeonPRO w7900 which has significantly less memory (only 384GB), GH200 costs the same, consumes 3x less energy and has a higher performance.
                  ‚Äč

                  I'm pretty confident that the above statements are completely and ridiculously‚Äč false.

                  Comment


                  • #59
                    Originally posted by coder View Post
                    Of course, I'm sure you're an expert on video content production, as with so many other things.
                    If you go to the video help forums you will find dozens of encoding tests I have done.

                    And as i mentioned i used to do consulting work for a music video production company out of Florida, so am I an expert?

                    You will note that the tests were not done with live action footage, which is industry standard way of testing encoder quality, they were done with video game captures.

                    You will also note that the speed comparison was of a top end Raptor Lake vs an older NVIDIA card and the author completely ignored that consumer NVIDIA cars are capable of encoding two streams at full speed simultaneously.

                    If you want to see better encoding tests, look for MSU's annual encoding tests.

                    Comment


                    • #60
                      Originally posted by qarium View Post

                      this cpu has 72 cores and is slower than a AMD 64core epyc/threadripper and a threadripper of course has even faster singlecore speed.
                      fewer cores is always nice but only if the performance is higher. this 72core gets a bonus for not having hyperthreating.
                      but of course you can disable hyperthreating on the threadripper so.
                      What benchmarks are you looking at?

                      Comment

                      Working...
                      X