Stories
Slash Boxes
Comments

SoylentNews is people

posted by Fnord666 on Sunday April 18, @01:57AM   Printer-friendly [Skip to comment(s)]

Nvidia Enters The Arms Race With Homegrown "Grace" CPUs:

Everything that Nvidia and the market has learned about CPUs in the past decade can be brought to bear in Grace CPUs, and now Nvidia itself is big enough to be the primary contractor for exascale-class machines or at the very least supply the CPU and GPU compute as well as the networking and DPU offload engines to a supplier such as Hewlett Packard Enterprise, which really doesn't mind making custom variants of its "Shasta" EX supercomputers. And that is precisely what will be happening with the future "Alps" supercomputer going into the Swiss National Supercomputing Center [(CSCS)] in early 2023, featuring a hybrid compute module that has the homegrown Grace Arm server CPU very tightly couple with a future Nvidia GPU accelerator.

[...] CSCS is not alone in getting a Grace-Hopper supercomputer. Los Alamos National Laboratory, one of the Department of Energy national labs in the United States, will also be getting a Grace-Hopper system supercomputer system HPE in early 2023.

[...] Nvidia estimates that a Grace-Hopper system will offer 10X the performance on training natural language models, reducing it from one month to three days, and will allow for real-time inference on a 500 billion parameter model on a single node. And on the HPC front, computational fluid dynamics and weather applications are expected to see performance on the Grace-Hopper systems as well. Other HPC applications will no doubt follow.

See also: For Arm-Driven Supercomputing, Nvidia is Right on Time
Intel CEO: 'We Are Playing Offense, Not Defense' In Response To NVIDIA Grace ARM CPU Announcement For Servers


Original Submission

Related Stories

Arm Pioneer: Nvidia's Grace CPU is Proof That It Will 'Compete Unfairly' 17 comments

Arm Pioneer: Nvidia's Grace CPU Is Proof That It Will 'Compete Unfairly'

Arm pioneer Hermann Hauser has once again criticized Nvidia's plan to acquire the semiconductor design company, with The Telegraph reporting Sunday that he believes Nvidia is "clearly showing it will compete unfairly" if the deal is approved.

Hauser's concerns reportedly centered on the Grace processor Nvidia announced at GTC 2021. The company's first Arm-based CPU will connect to high-end GPUs via NVLink, which purportedly offers data transfer speeds up to 900 GBps. That's significantly faster than other technologies—it's also exclusively available to Nvidia.

This is why Hauser told The Telegraph that he believes using a proprietary interface like NVLink could end up "locking customers into [Nvidia] products," which "clearly shows that they will compete unfairly with other Arm-based server companies such as Amazon and Fujitsu," rather than retaining Arm's neutrality.

Previously:
Nvidia-Branded ARM CPUs; UK Trade Union Speaks Out Against Deal
Nvidia's $40 Billion ARM Acquisition: "All but Dead"?
Nvidia Enters the Arms Race with Homegrown "Grace" CPUs
UK Will Investigate Nvidia's $40 Billion Arm Acquisition Attempt


Original Submission

Display Options Threshold/Breakthrough Reply to Article Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(1)
  • (Score: 1, Interesting) by Anonymous Coward on Sunday April 18, @02:01AM (6 children)

    by Anonymous Coward on Sunday April 18, @02:01AM (#1138908)

    It's the apocalypse I tell you !

    Intel makes gpus, nvidia makes cpus...

    Dogs and cats, living together. Mass histeria.

    • (Score: 2) by fakefuck39 on Sunday April 18, @03:48AM (5 children)

      by fakefuck39 (6620) on Sunday April 18, @03:48AM (#1138928)

      >Intel makes gpus, nvidia makes cpus...

      which makes absolute perfect sense. Intel makes top of the line CPUs, and shitty GPUs. This is useful for general compute farms and business workstations. So if CPU is your main need, but you do need to be able to display an Oracle management GUI, or draw a network diagram in Visio - you're all set. They solder in a shittly little GPU next to their top of the line CPU, and you're good - instead of general interconnects and ports and support for 3rd party video cards.

      nvidia is for stuff that can use lots of parallel processing and matrix transformations, and they're great at that. things like crypto mining, high frequency trading, etc. I've built many of these - a GPU-heavy cluster wired to a mellanox, usually with an all-flash nas array in there. and I always need to throw in a couple of shitty last-gen xeons in there, because you need very basic program logic to execute, but that's not the bulk of your processing. so now, they take their GPUs, solder in a little shit-chip general arm processor in there for some basic program branching and control logic, and you're all set.

      so what was crazy was before. when 99% of your load is matrix transforms and because of the control flow that a microwave CPU can handle, you gotta add support for a xeon and all those complexities. this here, with both intel and nvidia, makes perfect sense.

      • (Score: 0) by Anonymous Coward on Sunday April 18, @12:17PM (4 children)

        by Anonymous Coward on Sunday April 18, @12:17PM (#1138990)

        > Intel makes top of the line CPUs
        I'm sorry, are you from the past?

        • (Score: 2) by fakefuck39 on Sunday April 18, @05:49PM (3 children)

          by fakefuck39 (6620) on Sunday April 18, @05:49PM (#1139054)

          No, I'm from the real world with datacenters and large server farms, not your desktop PC that you play games on. In that world, AMD beats Intel because they push the latest fab out the door for datacenter, immediately, while Intel puts it on consumer products first, works out the bugs, and only then puts it on the Xeon line. I'm also from the world that builds quotes and configs for datacenters - and maybe 5% of that is AMD, the 95% being Intel. You are comparing last-gen fab for Intel against bleeding edge AMD. Bleeding edge is not what companies want to run production on. So yes, Intel makes top of the line CPUs, and AMD is not even a competitor. Any more questions about the real world before you go back to your basement and your computer games?

          • (Score: 0) by Anonymous Coward on Sunday April 18, @07:29PM (2 children)

            by Anonymous Coward on Sunday April 18, @07:29PM (#1139086)

            Intel didn't use last-gen fabs because datacenters and server farms want stability. They did it because they stumbled at 10nm for years. Intel went from bleeding edge to bleeding market share. Intel's CPUs have more bugs than AMD too. Companies will use Intel anyway because they can guarantee a consistent supply of chips.

            • (Score: 2) by fakefuck39 on Sunday April 18, @09:08PM (1 child)

              by fakefuck39 (6620) on Sunday April 18, @09:08PM (#1139124)

              As someone who literally designs and sells this for datacenters of large global enterprises, yous little theory is simply wrong. I've never had a customer ask me for an Intel quote because "Intel can guarantee supply." I can sell some vsan nodes with Intel or AMD. AMD uses a fab that came out 2 months ago, Intel's has been running on desktop for over a year. No bank is putting their financials on AMD, period.

              Bugs - like spectre? They're not bugs. Bugs make 2+2=5. Bugs make your database corrupt data and crash. Intel's bugs means performance is reduced, and you have to simply add some more cores. No one cares - cores are dirt cheap. Your ESX farm costs $1mil, you add another 50k in CPUs. You have data corruption or a miscalculation or an outage, you lose tens of millions. AMD has actual hardware errors resulting in failures and outages. Buying AMD for your prod environment is a good way to get fired.

              You know why they "stumbled" with 10nm for a long time? Because they didn't want to release anything buggy. Then they stumbled some more using it on consumer CPUs. Then finally, once it was solid and proven, did it make it into Xeon. This is called reliability - the opposite of what AMD delivers.

              • (Score: 0) by Anonymous Coward on Monday April 19, @02:08PM

                by Anonymous Coward on Monday April 19, @02:08PM (#1139374)

                [quote] AMD has actual hardware errors resulting in failures and outages. Buying AMD for your prod environment is a good way to get fired.[/quote]
                [Citation Needed]

                Oh look, a modern Intel chip failing in production at on of the biggest cloud vendors: https://blog.cloudflare.com/however-improbable-the-story-of-a-processor-bug/ [cloudflare.com]

  • (Score: -1, Troll) by Anonymous Coward on Sunday April 18, @02:21AM (4 children)

    by Anonymous Coward on Sunday April 18, @02:21AM (#1138912)

    WTF is this grace CPU?

    Man, I hate this millenial speak.

    • (Score: 1, Touché) by Anonymous Coward on Sunday April 18, @02:33AM

      by Anonymous Coward on Sunday April 18, @02:33AM (#1138914)

      Poor, poor delicate daffodil! It is so hurtful and offensive hearing a word one doesn't know!

    • (Score: 1, Interesting) by Anonymous Coward on Sunday April 18, @03:16AM

      by Anonymous Coward on Sunday April 18, @03:16AM (#1138922)

      "Grace" is the name for their new ARM processor line. I had to look at the article to find out what "Hopper" means. It is the Nvidia GPU that goes with their "Grace" CPU. Which is a rebranded ARM. The rest of the article looks like marketing drivel. TL;DR.

    • (Score: 1, Informative) by Anonymous Coward on Sunday April 18, @07:55AM (1 child)

      by Anonymous Coward on Sunday April 18, @07:55AM (#1138959)

      https://en.wikipedia.org/wiki/Grace_Hopper [wikipedia.org]

      smart and funny computing pioneer: https://www.youtube.com/watch?v=lGTEUtS5H7I [youtube.com]

      • (Score: 0) by Anonymous Coward on Sunday April 18, @05:47PM

        by Anonymous Coward on Sunday April 18, @05:47PM (#1139053)

        Hey that's my porn name.

  • (Score: 0) by Anonymous Coward on Sunday April 18, @11:33AM

    by Anonymous Coward on Sunday April 18, @11:33AM (#1138980)

    Nvidia has been designing and selling in-house ARM cores for a while now. Tegra stated in 2008 with Cortex cores but 2014 brought their own core - Denver, which was a really interesting design.

(1)