As we wrote about in September of last year, the supercomputing world was being dominated by Chinese entrants. For the first time in 5 years, the U.S. no longer sported a top 3 fastest supercomputer. The Titan supercomputer at Oak Ridge National Laboratory clocked a measly 17.6 petaflops, running up against the top two Chinese systems, clocking in at 93 and 33.8 petaflops respectively. And those differences aren’t just numerical bragging rights — they represent the ability to blaze the trail on AI, the most important technology in the future of computing. Well, Summit is here, and America finds itself once again with supercomputing supremacy.
Summit is also at Oak Ridge, but unlike Titan, its stated top speed is 200 petaflops, more than 10x faster than our previous best. It hasn’t been fully unleashed just yet, but that’s the theoretical peak for the new monster machine, which came on line officially this week.
To put that kind of speed in relative terms, Summit is calculating at such a rapid velocity that it’d take about 6.3 billion people performing a calculation every single second, together in unison, for an entire year to match what Summit can do in one second. ONE. Even that is such a large number it doesn’t do a great job of putting anything into relative perspective. The scope of these machines really is beyond comprehension at this point.
The Oak Ridge team says the $200 million system is the first supercomputer made specifically for use in artificial-intelligence applications. AI has been compared to the space race in numerous outlets — this one included — and there are striking number of germane similarities. For one, accomplishing the end goal requires a minimum threshold be met. For moon walking, it was minimum thrust vs. rocket weight for escaping earth’s atmosphere. For AI, it’s the size of the data set and the speed of the calculating machine. If you don’t have the raw computational horsepower, you’re operating at a disadvantage when it comes to AI. Another similarity is that getting to the destination is hardly the only goal — all the scientific advances made on the way to achieving the goal end up spawning nascent industries (GPS anyone?). So, it’s not just bragging rights at stake here — it could very well determine the future of our technological lives.
In building Summit to tackle AI specifically, the new system features a novel architecture from Titan, relying far more on GPUs from Nvidia to handle much of the computational workload as GPUs are often better matches for sifting through the massive datasets intrinsic to machine learning. Stacking GPUs to turbocharge complex computing systems mirrors what we’re seeing from most of the private sector leaders in artificial intelligence, machine learning and neural networks, which shouldn’t be all that surprising. Not all supercomputers use this build factor, called heterogeneous architecture, but the boost in chip performance is remarkable.
And what might that look like in practice? Well, according to the MIT Technology Review, “Among the AI-related projects slated to run on the new supercomputer is one that will crunch through huge volumes of written reports and medical images to try to identify possible relationships between genes and cancer. Another will try to identify genetic traits that could predispose people to opioid addiction and other afflictions.”
In the end, it’s not about bragging rights in and of themselves. The country controlling the biggest and fastest machine has a distinct advantage in the years to come. Not just in the heavy computational lifting, but also mission critical systems like cyber-security and the like. Summit is a massive and laudatory step forward, and it looks like Oak Ridge might just be getting started.