Announcement

Collapse
No announcement yet.

NVIDIA Announces Grace CPU For ARM-Based AI/HPC Processor

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

  • NVIDIA Announces Grace CPU For ARM-Based AI/HPC Processor

    Phoronix: NVIDIA Announces Grace CPU For ARM-Based AI/HPC Processor

    NVIDIA announced today in kicking off GTC21 the "Grace" high performance Arm processor for AI and high performance computing workloads. But before getting too excited, this high performance Arm chip isn't expected to be ready until 2023...

    https://www.phoronix.com/scan.php?pa...IDIA-Grace-CPU

  • #2
    CSCS = Swiss National Supercomputing Center, by the way.

    Comment


    • #3
      This is part of Nvidia's agreement to buy ARM. They stated they would build a state of the art AI Center in Europe to develop best in class ARM based AI and HPC solutions and to progress the whole AI eco-system with new tooling and programming advances.

      The Age of ARM is here. x86 is legacy.

      Comment


      • #4
        Oh..right...almost forgot. Not only will that Neoverse ARM SoC be tied to Nvidia GPUs through NVLink but the Grace cores will be ARMs Neoverse V1 cores like the Fujitsu Fugaku Supercomputer. Meaning that each core contains ARM's new SVE (Scalable Vector Extensions) that can range from 128 bit to 2084 bit in 128 bit chunks. You need vector routines in CPU? Check. You need more than Intel's pathetic 512 bit legacy AVX capabilities? Check. You need Massively Parallel capability like in a GPU? Just push off to the Nvidia GPU. Check. You need full cache coherency and zero memory copy with HBM on the SoC and HBM on the GPU? Check.

        x86 ?? Check...mate.

        Comment


        • #5
          Originally posted by Jumbotron View Post
          Oh..right...almost forgot. Not only will that Neoverse ARM SoC be tied to Nvidia GPUs through NVLink but the Grace cores will be ARMs Neoverse V1 cores like the Fujitsu Fugaku Supercomputer. Meaning that each core contains ARM's new SVE (Scalable Vector Extensions) that can range from 128 bit to 2084 bit in 128 bit chunks. You need vector routines in CPU? Check. You need more than Intel's pathetic 512 bit legacy AVX capabilities? Check. You need Massively Parallel capability like in a GPU? Just push off to the Nvidia GPU. Check? You need full cache coherency and zero memory copy with HBM on the SoC and HBM on the GPU? Check?

          x86 ?? Check...mate.
          Fugaku (A64FX) doesn't use a Neoverse-V1 but rather a Fujitsu custom core related to the SPARC and mainframe cores they already had inhouse. It is most closely related to the SPARC64 XIfx that preceded it in the FX100 supercomputer.

          Nvidia Grace will be using a future Neoverse core; there's no indication at this time that it uses V1.

          Comment


          • #6
            Originally posted by Procyon123 View Post
            CSCS = Swiss National Supercomputing Center, by the way.
            HP is building both Supers for the Swiss and Los Alamos.

            Comment


            • #7
              Originally posted by Dawn View Post

              Fugaku (A64FX) doesn't use a Neoverse-V1 but rather Zeus)jitsu custom core related toZeus)!) ) SPARC and mainframe cores they already had inhouse. It is most closely related to the SPARC64 XIfx that preceded it in the FX100 supercomputer.

              Nvidia Grace will be using a future Neoverse core; there's no indication at this time that it uses V1.
              " The new SVE features for the V1 Neoverse core draw on Arm’s experience working with Fujitsu on the A64FX platform, the SVE-enabled processor that is at the heart of Fugaku, the world’s top-ranked and first #1 Arm-based supercomputer. "

              " Silicon partners will have full control over SVE voltage and frequency transitions, said Bergey. This enables them to run at full frequency while executing SVE code, as Fujitsu’s A64FX CPU is able to do. "

              https://www.hpcwire.com/2020/09/22/a...rse-platforms/

              The Grace SoC will be V1 (Zeus) as N2 (Perseus) will be more for massive scale out and for TDP restricted use cases as Edge compute nodes. It could be Poseidon (V2 / N3) but ARM states that won't be out until late 2022 or 2023. Grace has to be engineered, tested and verified before then. So...V1 it is.

              Comment


              • #8
                Originally posted by Jumbotron View Post

                " The new SVE features for the V1 Neoverse core draw on Arm’s experience working with Fujitsu on the A64FX platform, the SVE-enabled processor that is at the heart of Fugaku, the world’s top-ranked and first #1 Arm-based supercomputer. "

                " Silicon partners will have full control over SVE voltage and frequency transitions, said Bergey. This enables them to run at full frequency while executing SVE code, as Fujitsu’s A64FX CPU is able to do. "

                https://www.hpcwire.com/2020/09/22/a...rse-platforms/

                The Grace SoC will be V1 (Zeus) as N2 (Perseus) will be more for massive scale out and for TDP restricted use cases as Edge compute nodes. It could be Poseidon (V2 / N3) but ARM states that won't be out until late 2022 or 2023. Grace has to be engineered, tested and verified before then. So...V1 it is.
                Yes. A64FX has SVE. A64FX is not a V1. It does not even resemble V1 microarchitecturally, which you would know if you looked closely at either; it resembles an XIfx with somewhat widened execution resources. It's not even an ARM Ltd designed core.

                Comment


                • #9
                  Originally posted by Jumbotron View Post
                  This is part of Nvidia's agreement to buy ARM. They stated they would build a state of the art AI Center in Europe to develop best in class ARM based AI and HPC solutions and to progress the whole AI eco-system with new tooling and programming advances.

                  The Age of ARM is here. x86 is legacy.
                  i dont like Nvidia but competition is always welcome... in the time of for the same performance 8000€ for intel 7800€ for AMD and 4000€ for a ARM cpu... this is really what we need to bring competition in the market.

                  but why only in the server market?... i would like to see a 128core ARM system for the desktop or workstation market.

                  also i think RISC-V and OpenPOWER is also the future.
                  Phantom circuit Sequence Reducer Dyslexia

                  Comment


                  • #10
                    Originally posted by Jumbotron View Post
                    Oh..right...almost forgot. Not only will that Neoverse ARM SoC be tied to Nvidia GPUs through NVLink but the Grace cores will be ARMs Neoverse V1 cores like the Fujitsu Fugaku Supercomputer. Meaning that each core contains ARM's new SVE (Scalable Vector Extensions) that can range from 128 bit to 2084 bit in 128 bit chunks. You need vector routines in CPU? Check. You need more than Intel's pathetic 512 bit legacy AVX capabilities? Check. You need Massively Parallel capability like in a GPU? Just push off to the Nvidia GPU. Check. You need full cache coherency and zero memory copy with HBM on the SoC and HBM on the GPU? Check.

                    x86 ?? Check...mate.
                    sounds really cool. can someone use a AMD MI100 CDNA instead of the nvidia bullshit GPU ?
                    Phantom circuit Sequence Reducer Dyslexia

                    Comment

                    Working...
                    X