China Tops Super Computer List, US Ousted Out of Top 3

The Piz Daint supercomputer. Placed third on the latest Top500 list.

TOP 500, the organization that monitors the performance and development of the world’s most powerful supercomputers, has released the

49^th edition of its “Top 500” list.

Unsurprisingly, China claimed the top two spots on the prestigious computing list with its Sunway TaihuLight receiving the highest honors, followed by the venerable Tianhe-2 (Milky Way-2). Last among the top three positions was the rather awkwardly named Swiss system “Piz Daint.”The United States claimed the fourth, fifth and sixth spots on the list.

The most remarkable aspect, however, isn’t the ordering of the list, but the large gulf in performance that exists between the first and second place systems. The Sunway TaihuLight, built by China’s National Research Center of Parallel Computer Engineering & Technology (NRCPC) and housed at the National Supercomputing Center in Wuxi, maxed out its potential with a Linpack rating of 93 petaflops. The next most powerful system – Cuangzhou’s National Supercomputer Center’s Milky Way-2– clocked in at 33.9 petaflops, nearly 1/3^rd as powerful as the world’s highest performance system.

Moving down to the third position, the Swiss National Supercomputing Centre’s Piz Daint made the most of its recent NVIDIA Tesla P1000 GPU upgrade, which increased its Linpack performance from 9.8 to 19.6 petaflops.

The 49^th installment of the TOP 500 list marks the second time in its 24-year history that the United States hasn’t had a machine in the top three tier of supercomputer rankings. That said, the country does hold the lion’s share of supercomputers on the list, placing 169 machines within the Top 500 performers.

Ultimately, what’s becoming apparent is that China is taking the development of its supercomputing capacity seriously, listing 160 systems among the Top 500.

Today, supercomputers are being used to develop weapon systems, model nuclear reactions, analyze complex meteorological data sets and much more. In the coming years, the US (and almost certainly China) will have their minds set to moving beyond the petaflop scale of supercomputing and into the exascale region of performance.

Exascale machines would be capable of performing one billion billion calculations per second, the equivalent of one thousand petaflops per second. To reach this level of performance, the US Department of Energy (DOE) has invested $258M in six companies to deliver an exaflop system within the 2021-2023 time frame. That kind of explosion in performance would be astonishing, and might transform the type of simulations that can be realistically executed on a supercomputing system, and how those simulations are programmed in the first place.