back to article Nvidia's Tesla P100 has 15 billion transistors, 21TFLOPS

Nvidia has revealed its Pascal Tesla P100 GPU, described as the largest FinFET chip on the market. It is in volume production today. It features 16nm FinFETs and 15 billion transistors on a 600mm2 die, 16GB of HBM2 memory, and the NVLink interconnect. If you include the memories, the 15 billion figure balloons to 150 billion …

  1. Roo
    Windows

    I think it's reasonable to quote single precision (32bit) FLOPS, given the origin of the term - but quoting half-precision figures is taking the biscuit. Presumably NVidia's benchmarketeers will be quoting quarter-precision FLOPS next time round to fool people into thinking their next gen is twice as quick as the P100.

    If pressed I would speculate that some real-time signal processing app out there can make use of 21 Thalflops, I'd be interested to hear what kind of apps folks think those halfFLOPs will be good for. :)

    1. bazza Silver badge

      @ Roo,

      "I think it's reasonable to quote single precision (32bit) FLOPS, given the origin of the term - but quoting half-precision figures is taking the biscuit."

      You beat me to it!

      "If pressed I would speculate that some real-time signal processing app out there can make use of 21 Thalflops, I'd be interested to hear what kind of apps folks think those halfFLOPs will be good for. :)

      Not a lot to be honest. To make good use of this chip one would have to load it up with a good chunk of data and then perform a whole cart load of sums on it. Otherwise one would simply be wasting time doing nothing but tiresome transfers across the PCIe bus.

      However, the more sums performed, the more significant the arithmetic errors will become as 16 bit half-floats struggle and fail to keep up with value growth that happens in, for example, an FFT. No matter which way one spins it, 16 bits can represent only 65536 different values; it's a bit like Trolls counting in Terry Pratchett books - "one, two, many, lots".

      But the 32 bit floats, yeah baby!

      I'm still not convinced though - a stonking great compute engine at the end of a PCIe connection is in the wrong place; you still have to transfer the data over to it to get sums done. If one's application doesn't need that many sums done on the data it'd simply be a waste of transistors.

      Certainly for some applications I can see the up-coming (or here already?) Xeon-Phi (the one that is a CPU in its own right) beating this chip, despite it maxing out at 6TFLOPs-ish, simply because the data and compute are already in the same place.

      China

      I expect the US government to be keen to not let China get hold of any of these. I read Intel aren't allowed to sell Xeon Phi to the Chinese, so letting NVidia sell this GPU to them would be a bit inconsistent. I don't know enough about the corporate structure of NVidia to know whether Uncle Sam has the same level of influence as they have over Intel.

      1. Cesar Maciel

        This adapter is based on NVLink, not PCI-e.

        http://www.nvidia.com/object/nvlink.html

  2. Swarthy Silver badge

    Can I get this GPU in a graphics card? I want to use it to finally try Crysis.

  3. kryptylomese

    From a domestic perspective, virtual reality will be calculated in real-time to look more realistic with this kind of computing power. Video could be processed to clean it up and so could audio.

    From a commercial perspective, farms of these will make movies in 3D and it could also be used to model weather systems

    From a criminal perspective, hacking passwords and keys.

    From a government perspective, hacking, nuclear simulation, weather and other modeling.

  4. werdsmith Silver badge

    15 billion transistors. I can't imagine what Geoffrey Dummer or Jack Kilby would think about it.

    How do engineers design circuits with 15 billion transistors?

    1. Anonymous Coward
      Anonymous Coward

      "How do engineers design circuits with 15 billion transistors?"

      By using computers with CPUs using only 4 billion transistors? Rinse and repeat till done.

    2. Roo

      "How do engineers design circuits with 15 billion transistors?"

      Macros. ;)

  5. Pascal Monett Silver badge
    Coat

    I can't put my finger on it . . .

    but for some reason, I like this card.

    Must be because it has Tesla in the name.

    Yup, that must be it.

    1. JeffyPoooh Silver badge
      Pint

      Re: I can't put my finger on it . . .

      It's going to fall in love with a pigeon, and then mumble incoherently about somehow powering the entire world with several transmitting towers.

      On the other hand, it'll be the best on Earth at modeling AC circuits.

  6. Sgt_Oddball Silver badge

    how big a psu would one of these need?

    On that note I wonder if you can sli them so you could really silly with the figures.

    Also does this mean that tesla will take over from titan?

    1. JEF_UK

      Re: how big a psu would one of these need?

      Tesla product line is like a quadro with no display output, for the data center.

      The Titan is a big geforce, for gaming.

      Expect power envelopes to be similar to the last generation, HBM saves power. That means more can be dissipated by the GPU/Core.

      I'm Hopping for some Geforce pascal news

      edit: apparently the TDP is 300W for the P100

      1. Updraft102 Silver badge

        Re: how big a psu would one of these need?

        Hop all you want if you think it will help!

  7. Jimbo in Thailand
    Happy

    16nm... Happy Days!

    With this announcement I'm guessing Nvidia's next-gen graphics cards (to be released June or July?) will also be 16nm. And AMD has also been hinting for months that their next-gen Radeons will be on a smaller die process. This is great news for us. Finally, rebranded/slightly-tweaked long in the tooth 28nm video cards will be relegated to the tar pits!

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019