China Muscles Out US on List of Fastest Supercomputers
Source: Jeffrey Burt
A new powerful system from China sits at the top of the Top500 list, and China is now home to more of those supercomputers than the United States.
China is now the world's dominant player in the global supercomputing space.
The country kept is position atop the Top500 list of the world's fastest supercomputers with a new system that not only is markedly faster and more efficient than the previous number-one supercomputer—also from China—but also is powered by processors designed and manufactured in China.
In addition, China now has more supercomputers on the list—167—than the United States, with 165, the first time in the more than two decades that the list has been kept that the United States is not home to the largest number of computers.
The latest list was released June 20 at the ISC High Performance 2016 show in Frankfurt, Germany. The Top500 list is released twice every year.
China's ascension comes less than two months after a report by the Information Technology and Innovation Foundation (ITIF) warned that the United States was in danger of losing its edge in the high-performance computing (HPC) field due to fast-growing competition primarily from China. In its report, the ITIF also outlined steps Congress needs to make—from increasing funding to amending export control regulations—to keep the country at the forefront of HPC and supercomputing development.
"Recognizing that both the development and use of high-performance computing are vital for countries' economic competitiveness and innovation potential, an increasing number of countries have made significant investments and implemented holistic strategies to position themselves at the forefront of the competition for global HPC leadership," report authors Stephen Ezell and Robert Atkinson wrote. "The report details how China, the European Union, Japan, and other nations have articulated national supercomputing strategies and announced significant investments in high-performance computing."
President Obama last year issued an executive order aimed at accelerating the development of high-performance computing systems in the United States.
The United States historically has been the dominant player in the space, though its position has eroded over the past several years. In the last Top500 list released in November 2015, the United States had 200 systems on the list. However, it was down from the 231 on the list released in July 2015. China, meanwhile, placed on 109 systems on the list, almost three times the country had on the July list.
China has been aggressive not only in developing faster supercomputers, but also creating the technologies that can be used in them. The new number-one system on the list, Sunway TaihuLight, is an example of this. The massive system provides 93 petaflops/second (quadrillions of calculations per second) of performance, and is more than twice as fast and three times as efficient as Tianhe-2, the Chinese system it displaced after being at the top of the previous six Top500 lists. Tianhe-2, powered by Intel's 12-core Xeon E5-2692 processors and using Chinese interconnect chips, offers 33.86 petaflops/second of performance.
Sunway TaihuLight was built by the National Research Center of Parallel Computer Engineering and Technology (NRCPC) and installed at the National Supercomputing Center in China. The system uses 40,960 Sunway SW26010 processors that hold more than 10.6 million cores. It's five times faster than Titan, a system at the Oak Ridge National Laboratory based on Cray's XK7 systems powered by AMD's Opteron chips. Titan is the fastest U.S. supercomputer on the list, coming in at number three.
The new Chinese system also is among the most efficient, with a peak power consumption under load of 6 Gflops/watt, according to organizers of the Top500 list.
The United States is making a push to continuing competing in the supercomputing space, particularly as the field moves toward exascale computing early in the next decade. The Department of Energy in 2014 signed up IBM and GPU vendor Nvidia to build two 150-petaflop supercomputers—Summit, for the Oak Ridge facility, and Sierra, for the Lawrence Livermore National Lab. Those systems aren't due to go online until 2017 or 2018, and another supercomputer—Aurora—isn't scheduled until 2019.
Other than the top spot, not much has changed among the top 10 fastest systems. The introduction of Sunway TaihuLight knocked off the Dell-based Stampede system at the Texas Advanced Computing Center (TACC), which was in the 10th slot on the November list. TACC officials earlier this month announced a $30 million award from the National Science Foundation to build a new large-scale supercomputer—Stampede 2—that will deliver twice the peak performance, memory, storage capacity and bandwidth of the current system, which went online in 2013. It will be based on Dell PowerEdge servers and powered by a mix of the next-generation of Intel Xeon chips and the chip maker's many-core Xeon Phi "Knights Landing" processors. It will be in place in 2018, providing 18 petaflops of peak performance.
Among other regions, Europe had 105 systems on the latest list, two fewer than in November. Asia in all had 218 systems on the list, up from 173.
There also is a continued slowdown in the total combined performance of all 500 systems. It's grown to 566.7 petaflops/second—over 420 six months ago. But the growth is smaller than the long-term trend had been, organizers said.
Intel continues to be the top chip provider in the list, with 455 systems, or about 91 percent. IBM's Power chips are in 23 systems, down from 26 in November, and AMD's Opteron chips are in 13 system, or 2.6 percent. AMD had 4.2 percent of the systems on the previous list.
Ninety-three systems on the list use accelerators or coprocessors to improve workload performance while keeping a lid on power consumption. Sixty-three of them use Nvidia GPUs and 26 use Intel's Xeon Phi coprocessors. Three use AMD's Radeon GPUs and two use Pezy custom chips from Japanese vendor Pezy Computing. Three of the systems use both Nvidia GPUs and Intel's Xeon Phis.
| }
|