China is the story coming out of the latest version of the twice-yearly Top500 list of the world’s fastest supercomputers, not only keeping hold of the No. 1 spot on the list for the sixth consecutive time, but also tripling the number of systems on the list overall.
There isn’t much change at the top of the list—there are only two new entrants in the top 10—and the overall performance of all systems on the list continues to slow, according to organizers of the list, the latest of which was released Nov. 16 during the SC 15 supercomputing show in Austin, Texas.
However, there is a lot of movement among the regions represented on the list, highlighted by China’s rapid growth and the decline in the number of systems on the list from the United States and Europe. China’s Tianhe-2 “Milky Way” supercomputer, which is housed at the country’s National University of Defense Technology, is still the top system—it’s been in the top spot since 2013, with a performance of 33.86 petaflops (quadrillions of calculations per second). The system, which almost doubles the performance of Titan, the system ranked second with a performance of almost 17.6 petaflops. Titan is housed at the Oak Ridge National Laboratories in the United States.
Tianhe-2 is powered by Intel’s 12-core Xeon E5-2692 processors and Xeon Phi coprocessors. Titan is based on Cray’s XK7 system, which runs on 16-core Opteron 6274 processors from Advanced Micro Devices and which uses Nvidia’s Tesla K20x GPU accelerator.
However, it was what China did down the list that is most impressive. In the previous list released in July, there were 37 supercomputers from the country. In the most recent list, that number went up to 109, surpassing the 108 in Europe (it was 141 in the list in July). The United States also is continuing to see its numbers dwindle, down to 200 on the new list. In July, there were 231 U.S. computers on the list, and the 200 represents the fewest number of U.S. systems since the listed was started in 1993.
Japan saw its numbers fall slightly, from 40 supercomputers in July to 36 in November.
Chinese system makers also are making their presence felt. Lenovo had three systems on the July list, but with IBM’s x86 server business securely in its pocket, Lenovo saw its number of systems on the list grow to 25. Top500 organizers noted that there are nine systems listed as IBM/Lenovo, and another five as Lenovo/IBM. Chinese vendor Sugon now has 49 systems on the list, passing IBM in the category.
China is pushing system makers to put more components made by Chinese companies into their servers and components. It’s part of the country’s effort to lessen the influence of U.S. OEMs like Hewlett Packard Enterprise and Dell and component makers like Intel in the tech environment in China.
However, China’s position at the No. 1 spot on the list may not last much longer. Other countries and vendors are targeting Tianhe-2. The U.S. Department of Energy (DoE) has awarded hundreds of millions of dollars to companies like IBM, Nvidia, Cray and Mellanox to build supercomputers that will be five to 10 times faster than Tianhe-2. The projects are part of the DoE’s FastForward 2 program aimed at accelerating the drive toward reaching exascale computing in five to six years, where systems will be 20 to 40 times faster than current supercomputers.
Cray has the most systems on the list, with 24.9 percent market share, officials said. IBM comes next at 14.9 percent, followed Hewlett Packard Enterprise. China has about 9.2 percent of the total performance on the list.
The two new entrants in the top 10 list are Trinity, a system coming in at No. 6 and based on Cray’s XC40 servers powered by 16-core XeonE5-2698v3. It offers a performance of 8.1 petaflops and is being run by the U.S. National Los Alamos National Laboratory and Sandia National Laboratories. At No. 8 is Hazen-Hen, a Cray XC40 systems running on Intel’s 12-core Xeon E5-2680 chips and running at the Höchstleistungsrechenzentrum Stuttgart HPC center. It has hit 5.6 petaflops of performance.
Other notable trends on the list are the continued slowing of the overall performance from all 500 systems. Over the past year, the total combined performance grew from 309 petaflops to 420 petaflops. In addition, the number of systems with a performance of more than a petaflop of performance grew from 67 in July to 80 now, and the number of systems using accelerators—such GPUs from Nvidia or AMD or x86 Xeon Phi coprocessors from Intel—has jumped from 90 in July to 104 in November.