Quantcast

During a keynote speech at our annual GPU Technology Conference in San Jose, Calif., NVIDIA CEO Jen-Hsun Huang updated our public GPU roadmap with the announcement of Pascal, the GPU family that will follow this year’s Maxwell GPUs.

Pascal module

Named for 17th century French mathematician Blaise Pascal, our next-generation family of GPUs will include three key new features: stacked DRAM, unified memory, and NVLink.

  • 3D Memory: Stacks DRAM chips into dense modules with wide interfaces, and brings them inside the same package as the GPU. This lets GPUs get data from memory more quickly – boosting throughput and efficiency – allowing us to build more compact GPUs that put more power into smaller devices. The result: several times greater bandwidth, more than twice the memory capacity and quadrupled energy efficiency.
  • Unified Memory: This will make building applications that take advantage of what both GPUs and CPUs can do quicker and easier by allowing the CPU to access the GPU’s memory, and the GPU to access the CPU’s memory, so developers don’t have to allocate resources between the two.
  • NVLink: Today’s computers are constrained by the speed at which data can move between the CPU and GPU. NVLink puts a fatter pipe between the CPU and GPU, allowing data to flow at more than 80GB per second, compared to the 16GB per second available now.
  • Pascal Module: NVIDIA has designed a module to house Pascal GPUs with NVLink. At one-third the size of the standard boards used today, they’ll put the power of GPUs into more compact form factors than ever before.

Pascal is due in 2016.

For more details on Pascal and Stacked Memory see this Parallel Forall post by NVIDIA Senior Director of Architecture Denis Foley.

  • symbolset

    If you can’t get more Moore in X and Y, use Z.

  • Grahaman27

    “NVIDIA has designed a module to house Pascal GPUs with NVLink. At one-third the size of the standard boards used today”

    that sounds really cool. the less the space it takes up the better.

  • Alfredo Rossi

    And the next one will be Newton with optical parts

  • https://www.youtube.com/channel/UCWryXRcB2hOTLqXXPs7MkBw Christian L

    Sounds expensive, still looking forward to it though. 2016 looks exiting already.

  • Yeltnerb1

    Maxwell is officially D.O.A

  • nilsm

    That would also mean we need a CPU that specifically supports NVLink, does it not? Or how would it connect to the CPU?

  • Yaz Akiera

    and also a motherboard to support the new interface.

  • Sam Tosi

    the “wafer” architecture design sounds good but it also sounds like Dram would get really hot compacted like that are we all going to have to switch to water cooling?

  • Rory Picard

    “and quadrupled energy efficiency.” less energy used means less heat.

  • John Smith

    From my understanding, dropping energy consumption by three quarters means heat output dropping by much more than 75%, i.e. the relationship is not linear.
    Bear in mind though that the DRAM could easily consume more energy overall as there is more of it.

  • Sam Tosi

    i forgot about the efficiency increase which could easily be a solution to a quite obvious problem. Thanks for that reminder Rory. On the other hand John The possibility that the added amount of memory could easily consume more energy is something to remember. However there is also the fact that you would only be using the amount of memory and energy allocated to the program so you won’t be using it all at once. So really it wouldn’t be any different in the heat department from any other card with similar capability except for the fact that the portion of the memory you would be using in the new architecture is sitting under other modules which are absorbing heat and trapping it in the card unlike your Dram modules are now. It will be interesting to see how it plays out. If it is a problem i know Nvidia will have a solution at some point

  • Sam Tosi

    they did say that it was not as big as the normal pcie slot correct? im going to be really upset if i have to upgrade my board in the next couple years

  • Generalkidd

    Why can’t NVLINK be used as the next USB standard or something like that? I would love to have an NVlink flash drive. Transfer rates of 80 GB/s? That’s pretty amazing!

  • Ed Noneofyourbiz

    Clever. Any Moore comments (aside from the one)?

  • Ed Noneofyourbiz

    Wouldn’t NVLink require CPUs and chipsets (motherboards) to be designed to support it?
    Might well be worth it.
    Computing power is becoming Moore awesome than ever.

  • Ed Noneofyourbiz

    I’d like to take about 100 top-end 2017 machines, with 64GB DDR4 RAM, Skylake (or Cannonlake) processors, top-end NVIdia Pascal GPUs, large SSD and hard storage, and suitable monitors, cables, keyboards, mice, etc., loaded with Linux, and every bit of source code I could find, into a time-machine, and come out about 1950.
    Maybe I could offer one to England in exchange for not (chemically) castrating Alan Turing.

  • Jeff Anthony S Enad

    I was impress of this Pascal. It would be great if there’s public test.

  • http://www.silver3d.it Federico

    Any relase date for the new maxwell quadro?