500 trillion int4 tops and 10 grand the new RTX 8000 from Nvidia

Discussion in 'Machine Learning, Deep Learning, and AI' started by gigatexal, Aug 13, 2018.

  1. gigatexal

    gigatexal I'm here to learn

    Joined:
    Nov 25, 2012
    Messages:
    2,484
    Likes Received:
    440
    when you don’t have any real competition (I think Nvidia and CUDA are not matched by anything) you get impressive hardware but at ridiculous prices. Buy a nice used car or a new professional GPU... the next Pixar movie should be amazing though.

    https://www.anandtech.com/show/13214/nvidia-reveals-next-gen-turing-gpu-architecture

    Next flagship quadro card among others were announced. The Quadro RTX 8000 had 48GB of GDDR6, 4608 CUDA cores, 576 tensor cores, can do 500 trillion “tops” at int4 precision etc etc.
     
    #1
    vv111y likes this.
  2. Patrick

    Patrick Administrator
    Staff Member

    Joined:
    Dec 21, 2010
    Messages:
    11,160
    Likes Received:
    4,118
  3. Robert Fontaine

    Robert Fontaine Active Member

    Joined:
    Jan 9, 2018
    Messages:
    112
    Likes Received:
    26
    I need a lot more money.
     
    #3
    vv111y likes this.
  4. ideabox

    ideabox Member

    Joined:
    Dec 11, 2016
    Messages:
    68
    Likes Received:
    23
    Only NVIDIA can make there 10K GPU obsolete in 8 Months and get away with it.

    They do it way too often for my liking, Unfortunately i still pay :/ but the space is prime for Competition. Looking forward to Graphcore or similar to come to life and give them a kick in the bum.
     
    #4
  5. Jim Dowling

    Jim Dowling New Member

    Joined:
    Aug 14, 2017
    Messages:
    4
    Likes Received:
    2
    The Vega 20 will be the really interesting card that comes out this year. They are already catching up with ROCm as a viable competitor to cuda. They have it working with the latest version of TensorFlow (1.8, i think - but they catch up quick) - they plan to try and get it merged to core, would be great if they could.
    The real question is what the Vega 20 will retail for?
     
    #5
    gigatexal likes this.
  6. gigatexal

    gigatexal I'm here to learn

    Joined:
    Nov 25, 2012
    Messages:
    2,484
    Likes Received:
    440
    STH says ROCm might be too little
    Too late tho
     
    #6
  7. Jim Dowling

    Jim Dowling New Member

    Joined:
    Aug 14, 2017
    Messages:
    4
    Likes Received:
    2
    That STH AMD article felt like one friend telling another to get their crap together. They are doing it. All we need is great tensorflow support and Vega20 will be huge. It is getting there. Their use of code generating compilers is really promising.
     
    #7
    gigatexal likes this.

Share This Page