Author(s): Raghu Gopal
Last week, the US Department of Energy unveiled a supercomputer called Summit at the Oak Ridge National Laboratory in Tennessee. Summit is powerful enough to allow the country to recapture the number-one spot for top-performing supercomputers for the first time since 2013.
Summit was developed together with IBM and is powered by 9,216 of its central processing chips along with 27,648 graphics processors from Nvidia. Unlike other supercomputers, it was developed specifically for artificial intelligence, and is optimized for machine learning.
The machine, which was funded by the US government and cost more than $200 million to build, is capable of 200 quadrillion computations per second or 200 petaflops, making it eight times faster than the US' previous top supercomputer and twice as speedy as the world's current leader, which is located at China's National Supercomputing Center in Wuxi. To put it simply, Summit is 1 million times faster than a typical laptop.
Supercomputers can accelerate the development of technologies at the forefront of computing, such as artificial intelligence, and handle vast amounts of data. For example, Summit has begun processing data generated by the Million Veteran Program, a US research project that invites volunteers to grant researchers access to their health records, contribute blood tests for genetic analysis and answer questions about their lifestyle. To date, 675,000 US military veterans have joined the programme. As its name suggests, the goal is to reach 1 million volunteers by 2021.
Primarily, Summit could enable large breakthroughs in the use of artificial intelligence and machine learning in science and technology research fields. This means that supercomputers capable of running the world's most powerful machine-learning algorithms would become invaluable assets for the development and implementation of new scientific advancements. As the global race for computing power heats up, supercomputers are seen as a measure of a nation's technological leadership as well as its future potential. And in particular, countries that can take full advantage of artificial intelligence could achieve significant economic gains.
Summit is a beast of a computer, but there's a healthy competitive atmosphere to outperform the latest and greatest machine. The US, China, Japan, and the EU have all declared a goal to build the world's first "exascale" computer, that is, a machine capable of 1,000 petaflops or at least 1 billion billion calculations per second. This so-called exaflop computing is now the next big milestone in large-scale computing. According to Stephen Ezell, vice president for global innovation policy at the US Information Technology and Innovation Foundation, China may pip the US and achieve the landmark by 2020, making the former's current leadership in supercomputing short-lived. Indeed, the bigger prize over the next decade will be leadership in quantum computing — a race already underway.
Sign up to our free Daily Insight service here.