The Race Is On For The World's First Exaflop Supercomputer

By Kevin Carbotte
published

President Obama wants the United States to lead the charge in supercomputer power, and he has just signed an executive order launching an effort to ensure that happens.

The National Strategic Computer Initiative (NSCI), which was signed by President Barack Obama today, is an effort to build the world's first supercomputer with an exaflop of computational power. To put that into perspective, that is approximately 30 times the computing power of the most powerful supercomputers that exists today.

According to Nvidia, to achieve such a level of processing power would require as much as two gigawatts of electricity, which is roughly the output of the Hoover Dam. To actually achieve such a lofty goal, advancements in performance and power efficiency will be needed. Nvidia said its GPU technologies are poised to deliver this efficiency, delivering as much as 10 times the operations per unit of energy as compared to CPU cores.

President Obama's NSCI is an extension to existing plans (opens in new tab) announced back in November to create two pre-exascale systems being built by the Department of Energy.  These two projects are already underway, and the systems being built are expected to deliver between 100-150 petaflops each. That's merely a fraction of the performance demanded by the NSCI, and yet those are already multiple times faster than the supercomputers in use today. The fastest supercomputer in the world at present, China's Tianhe 2, is only capable of a peak 54.9 petaflops of computational power.

The United States isn't the only nation on a quest for exaflop computing power, but the executive order signed today is designed to enable the U.S. to be the first.

To aid in the development of such high performance parallel computing machines, Nvidia is developing a high speed interconnect called NVLink. The company has said that NVLink will enable CPUs and GPUs to exchange data at a much higher rate. Nvidia claimed it allows the CPU and GPU to communicate as much as 5 to 12 times faster than existing interconnect technologies.

Nvidia is also leveraging its recently released OpenACC toolkit (opens in new tab) to simplify use of these massively parallel computing devices. OpenACC makes it simple for researchers to write code that can be executed on different exascale architectures, thus making it easy to run the same code on vastly different machines without rewriting anything. Nvidia's key focus was productivity and portability when creating the toolkit.

Nividia said that scientists are already dreaming up new ways to make use of these forthcoming exascale computers. The European Commission's Human Brain Project plans to use exascale computing to better understand the human mind, and the White House hopes to speed the advancement of personalized medicine with its Precision medicine initiative.

With scientists, engineers and government thinking on an exaflop scale, computing power is about to take a massive leap forward.

Follow Kevin Carbotte @pumcypuhoy. Follow us @tomshardware, on Facebook and on Google+.

Kevin Carbotte

 Kevin Carbotte is a contributing writer for Tom's Hardware who primarily covers VR and AR hardware. He has been writing for us for more than four years. 

31 Comments Comment from the forums
  • Vlad Rose
    Skynet, here we come. :)
    Reply
  • nevilence
    Skynet, here we come. :)

    Come now, even the end of humanity needs a shinny toy
    Reply
  • So, humm, Nvidia is going to get a boatload of money to make a supercomputer we don't need?

    What a waste of money. I think I read that for 16 billion, you could reduce the cost of university by half. Instead, let's make a supercomputer that will be used to ego boosting!
    Reply
  • MasterMace
    Weather models are getting more accurate as the computers calculating them get more powerful. Picture a weather model 30 times more robust than what you have today. Superstorm Sandy was projected 7 days in advance. New England doesn't even bother with predictions further than 3 days due to inaccuracy. Figure that storms coming off Africa will have much higher predictions. If you can see a storm coming from 2 weeks out with a high degree of accuracy, that is extremely important.

    That's just your day to day life. Genetics is another thing requiring large amounts of computational power. This will help cure diseases, and discover ones that exist that we don't know about.
    Reply
  • jaber2
    We already have the CPU power, unfortunately this power is currently used it to take images of our private parts and send it to each other, if they could come up with a app that can use smartphones to execute a part of the program then we wouldn't need to build these.
    Reply
  • kcarbotte
    We already have the CPU power, unfortunately this power is currently used it to take images of our private parts and send it to each other, if they could come up with a app that can use smartphones to execute a part of the program then we wouldn't need to build these.

    I don't want my battery being drained even more than it already is. As it stands many modern smartphones can't make it through the day without a batter backup helping it along. No one wants thier phone being used when it's not them using it.

    To paraphrase something Neil Degrass-Tyson said, taking away the funding from NASA takes away the publics ability to dream of the future.
    Well I'd say this qualifies as a good stand-in.

    We need this kind of computational power to learn more about the human brain.
    We need this kind of computational power to simulate weather systems
    We need this kind of computational power to simulate deep space travel for manned and unmanned craft.
    We need this kind of computational power to simulate new energy production methods before actually attempting them.

    There's countless uses for supercomputers, that's why developped nations around the world have been building them for decades.
    Reply
  • emergent84
    No one in the world is going to build an exoscale SC without IBM, Intel, Nvidia processors, chip-sets, interconnects ,etc. Only China and the US will build one. And the US has somewhat attempted to crackdown on China Gov't using these. No doubt China has a workaround. It will be interesting to see what kind a shell company or University China has to acquire "legally" the massive amounts of US tech. I would like to see the US Gov't either buy up all the supplies, or an export ban for the next two gens of US HPC related components.

    Reply
  • dstarr3
    So, humm, Nvidia is going to get a boatload of money to make a supercomputer we don't need?

    What a waste of money. I think I read that for 16 billion, you could reduce the cost of university by half. Instead, let's make a supercomputer that will be used to ego boosting!

    You're right. Let's just stagnate and never progress or grow again.
    Reply
  • JPNpower
    We already have the CPU power, unfortunately this power is currently used it to take images of our private parts and send it to each other, if they could come up with a app that can use smartphones to execute a part of the program then we wouldn't need to build these.

    Umm... BOINC?
    or Folding@Home?
    Reply
  • falchard
    Man I hate Obama. Another executive order. He is acting more like a King than a President.
    Reply