NVIDIA Announces Grace CPU for Giant AI and High Performance Computing Workloads

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Apr 13, 2021.

  1. Hilbert Hagedoorn

    Hilbert Hagedoorn Don Vito Corleone Staff Member

    Messages:
    48,325
    Likes Received:
    18,407
    GPU:
    AMD | NVIDIA
    DeskStar and insp1re2600 like this.
  2. Lowice

    Lowice Active Member

    Messages:
    62
    Likes Received:
    19
    GPU:
    MSI 980ti OC
    Add it to the new Nvidia Shield TV and I'll buy it :)
     
    cucaulay malkin likes this.
  3. DeskStar

    DeskStar Guest

    Messages:
    1,307
    Likes Received:
    229
    GPU:
    EVGA 3080Ti/3090FTW
    Wholly WOWO!!!

    This is truly something to behold!!

    10,000 years of engineering!!! Amazing how we can use compute power of today in order to attain information of tomorrow!!

    Love to see which CPU Nvidia is using to only get 200GB/s of throughput?!?

    Ummmmmmmm..... Pretty sure my 3900X is capable of handling information in the realm of thousands of gigabytes per second?!?! [​IMG]

    Maybe I'm seeing something differently?!?

    Maybe?
     
    Last edited: Apr 13, 2021
  4. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    900 GB/s connection between Grace and NVIDIA GPUs to enable 30x higher aggregate bandwidth compared to today's leading servers
     
    DeskStar likes this.

  5. Denial

    Denial Ancient Guru

    Messages:
    14,201
    Likes Received:
    4,105
    GPU:
    EVGA RTX 3080
    They are talking about the bandwidth between interfaces, not cache bandwidth. Also why is your picture so massive.
     
    DeskStar and cucaulay malkin like this.
  6. Noisiv

    Noisiv Ancient Guru

    Messages:
    8,230
    Likes Received:
    1,494
    GPU:
    2070 Super

    I am not.

    The poor thing needs 4,000 pixels of height for a single Aida64 screenie.
     
    cucaulay malkin likes this.
  7. schmidtbag

    schmidtbag Ancient Guru

    Messages:
    7,955
    Likes Received:
    4,336
    GPU:
    HIS R9 290
    It's ARM - says so in the first sentence of the article.
    Even if we were to disregard the clarification made earlier, we're talking data centers here, so efficiency is often more important than total performance.
     
  8. cucaulay malkin

    cucaulay malkin Ancient Guru

    Messages:
    9,236
    Likes Received:
    5,208
    GPU:
    AD102/Navi21
    "but my Ryzen :mad::mad::mad::mad:"

    I'm sure this will be exceptional for very specific workloads only and just okay for anything else.
    you don't need it.
     
    Last edited: Apr 13, 2021
  9. Astyanax

    Astyanax Ancient Guru

    Messages:
    16,998
    Likes Received:
    7,340
    GPU:
    GTX 1080ti
    If you slice off the top section of the chip, you also get nvidia's next Tegra ;)

    [​IMG]
     
  10. DeskStar

    DeskStar Guest

    Messages:
    1,307
    Likes Received:
    229
    GPU:
    EVGA 3080Ti/3090FTW
    Your point? Obviously I was mistaken. What are you here contributing???!
     

  11. DeskStar

    DeskStar Guest

    Messages:
    1,307
    Likes Received:
    229
    GPU:
    EVGA 3080Ti/3090FTW
    Thank you. Like it says in my comment. "Maybe I missed something?"

    But asshats on here want to take it as if they're the engineer behind it all defending their precious it would seem.

    Comments are meant for clarification if need be not everyone is an engineer/professional on the subject.

    Seems as if guru3d is turning into an opinionated cesspool IMO....
     
    Last edited: Apr 14, 2021
  12. DeskStar

    DeskStar Guest

    Messages:
    1,307
    Likes Received:
    229
    GPU:
    EVGA 3080Ti/3090FTW
    "The Poor thing"??

    I'm not an imgur engineer, so I can't attest to how they format their uploads.

    And your comment is making light of mine how...? Other than being a placeholder for idiocy?

    Just wondering as all I stated was what I knew of the situation. And you're here with what?
     
  13. DeskStar

    DeskStar Guest

    Messages:
    1,307
    Likes Received:
    229
    GPU:
    EVGA 3080Ti/3090FTW
    Because that's the way imgur formatted it? Didn't care to crop it from my phone before uploading? Usually it's a link to click on, but yeah I don't know.
     
  14. Crazy Joe

    Crazy Joe Master Guru

    Messages:
    282
    Likes Received:
    124
    GPU:
    RTX 3090/24GB
    The mistake you are making is that you are comparing the cache performance of your CPU with the interconnect speed of the CPU to the GPU. Your Ryzen chip will only achieve maximally PCI-E 4.0 x16 speeds when communicating to a GPU (if that supports PCE-E 4.0 as well).

    At the moment that is maximally 32 GB/s (in one direction, but the bus supports simultaneous transfers in both directions at those speeds). There is most likely some overhead (I've tested an A100 on PCI-E 3.0, which could maximally achieve 16 GB/s, but that doesn't go beyond 13.2 GB/s), so expect a real world value in the region of 28-30 GB/s.

    So the 200 GB/s you are quoting is around 7x what your Ryzen can achieve. And in the slides that I saw they quoted a 500 GB/s number, so more like around 17x.
     
    DeskStar likes this.
  15. DeskStar

    DeskStar Guest

    Messages:
    1,307
    Likes Received:
    229
    GPU:
    EVGA 3080Ti/3090FTW
    And thank you for the information and your time it took you to share.

    Thank you so much.
     

  16. jura11

    jura11 Guest

    Messages:
    2,640
    Likes Received:
    707
    GPU:
    RTX 3090 NvLink
    Nvidia should include hashrate in marketing materials hahaha
     
  17. theoneofgod

    theoneofgod Ancient Guru

    Messages:
    4,677
    Likes Received:
    287
    GPU:
    RX 580 8GB
    Don't be giving them any ideas.
     

Share This Page