China's massive Intel-powered Tianhe-2 supercomputer—coming out two years ahead of schedule—is ending the Cray Titan's run as the fastest system in the world.
At the 2013 International Supercomputing Conference in Germany June 17, Tianhe-2—code-named Milky Way-2—sat atop the bi-annual Top500 list of the world's fastest supercomputer, with a performance of 33.86 petaflops per second. The move dropped Titan—a Cray system installed at the U.S. Department of Energy's (DOE's) Oak Ridge National Laboratory in Tennessee—to the No. 2 spot, ending Titan's run of seven months at the top of the list.
Tianhe-2's performance almost doubled the 17.59 petaflops per second of Titan.
The Top500 list also showed off a number of other trends, including Intel's dominance, the continued use of accelerators to speed up the performance of some supercomputers, the popularity of IBM's BlueGene/Q systems and the increasing performance of supercomputers in general.
Tianhe-2 also indicated the work the Chinese are doing in developing their own supercomputing technology. The system—which is arriving two years ahead of schedule and is being developed by China's national University of Defense Technology—is primarily based on technology developed in that country, according to Jack Dongarra, editor of the Top500 list, who toured the Tianhe-2 facility in May.
"Most of the features of the system were developed in China, and they are only using Intel for the main compute part," Dongarra said. "That is, the interconnect, operating system, front-end processors and software are mainly Chinese."
China has 66 systems on the Top500 list and is now the No. 2 in high-performance computing (HPC), behind the United States and ahead of Japan, the U.K., France and Germany. In addition, thanks to Tianhe-2, China also has leapfrogged Japan into the No. 2 position in overall compute performance, according to the list.
Tianhe-2, which will be deployed at China's National Supercomputer Center by the end of the year, has 16,000 compute nodes, each of which includes two 2.2GHz Xeon E5-2692 processors and three Xeon Phi coprocessors, for a combined 3.12 million computing cores. Tianhe-2's ascension marks the first time since November 2010—when the Tianhe-1 system was at the top of the list—that China has had the world's fastest supercomputer.
Titan is based on Cray's XK7 systems and is powered by Advanced Micro Devices' Opteron 6274 processors. It holds 560,640 processors, including 261,632 of Nvidia's Tesla K20x GPU accelerators. It's also among the most energy-efficient systems on the list, consuming 8.21 megawatts of power and delivering 2,143 megaflops of performance per watt, according to the Top500 list.
Rounding out the list of the top five fastest supercomputers are Sequioa, a system at the DOE's Lawrence Livermore National Laboratory that is based on IBM's BlueGene/Q and runs on more than 1.57 million Power cores, Fujitsu's K Computer—powered by 705,024 of the vendor's SPARC64 chips—at the RIKEN Advanced Institute for Computational Science in Japan, and another IBM BlueGene/Q system, Mira, at the Arronne National Laboratory, that is powered by 786,432 Power cores.
According to the Top500 list—compiled by Hans Meuer of the University of Mannheim in Germany, Erich Strohmaier and Horst Simon of Lawrence Berkeley National Laboratory, and Dongarra of the University of Tennessee in Knoxville—Intel continues to power most of the systems—80.4 percent—on the list.
In addition, Intel's Xeon Phi coprocessors, which officially rolled out last year, run in two of the top 10 fastest supercomputers—Tianhe-2 and the Dell PowerEdge-based Stampede system at the Texas Advanced Computing Center at the University of Texas in Austin—and in 11 systems on the entire list. In all, 54 systems on the list run accelerators, with 39 of those using Nvidia GPU chips and another three using AMD's ATI Radeon graphics technology. In the November 2012 list, there were 62 systems that used accelerators or coprocessors.
Twenty-six systems on the list offer a performance of more than a petaflop per second, a slight increase over the 23 from the November 2012 list. Eighty-eight percent of the systems use chips with six or more cores, and 67 percent use processors with eight or more.
IBM's BlueGene/Q is in four of the top 10 fastest systems.
The total combined performance of all 500 supercomputers on the list is now at 223 petaflops per second, compared with 123 petaflops per second a year ago. The last system on the list this year—the Intel-based Web Company system in the United States, at 96.6 teraflops per second of performance—was at No. 322 in the November 2012 list.
Of the 500 systems on the list, 252 are in the United States, while 119 are in Asia and 112 are in Europe.