SEATTLE – Fujitsu’s K Computer, which in June put Japan back at the top of the list of the world’s most powerful supercomputers, kept the No. 1 ranking in the latest list, becoming the first system to pass the 10-petaflop barrier.
The supercomputer, which is being installed at the Riken Advanced Institute for Computational Science (AICS) in Kobe, Japan, hit a peak performance of 10.51 petaflops (quadrillions of floating-point operations per second), according to the latest list released Nov. 14 by the Top500.org group.
In June, the K Computer, which is powered by 705,024 of Fujitsu’s SPARC64 processors spread over 864 server racks, hit a peak of 8.16 petaflops. It was the first time since 2004-when Fujitsu’s Earth Simulator was dethroned-that a Japanese computer led the list.
The list was released on the first day of the SC 11 supercomputing show here, which is expected to draw as many as 10,000 people.
The K Computer easily outdistanced the No. 2 system, the Tianhe-1A supercomputer in China, which hit a peak of 2.57 petaflops. A year ago, the Tianhe-1A system was on top of the Top500 list. Overall, the list released Nov. 14 looks exactly as it did in June.
“This is the first time since we began publishing the list back in 1993 that the top 10 systems showed no turnover,” TOP500 editor Erich Strohmaier said in a statement.
Rounding out the top five are a supercomputer at the Oak Ridge National Laboratory, powered by Cray’s XT5 systems, which run on Opteron processors from Advanced Micro Devices; a system at the National Supercomputer Center in China, powered by Intel’s Xeon processors and Nvidia’s graphics technology; and the system at the GSIC Center at the Tokyo Institute of Technology, which is based on Intel-based ProLiant servers from Hewlett-Packard that also leverage Nvidia GPUs.
Fujitsu’s K Computer doesn’t use any graphics accelerators, which are becoming more popular in helping systems makers increase performance as well as energy efficiency. According to the Top500 list, 39 systems now use such graphics accelerators, more than twice the 17 systems on the list six months ago. Thirty-five of those use Nvidia graphics chips, two use IBM’s Cell processor and two use AMD’s ATI Radeon GPUs. Three of the top five are using Nvidia GPUs to accelerate the computational capabilities, according to the list.
In addition, the core counts on the chips being used in these HPC systems are going up, according to the list. About 62 percent of all the supercomputers on the list use processors that offer six or more cores, and that is sure to go up. AMD on Nov. 14 officially launched its Opteron 6200 “Interlagos” processors, which offer up to 16 cores.
AMD rival Intel remains the top chip vendor to the HPC space, powering 76.8 percent-or 384-of the supercomputers in the Top500 list. However, that is a slight drop from six months ago, when Intel claimed 386 systems, or 77.2 percent. AMD is second with 63 systems, or 12 percent, according to the list. That also is down slightly, from 66 supercomputers on the June list.
IBM’s Power processors are No. 3, with 49 systems, or 9.8 percent, up from 45 supercomputers.
IBM is still by far the top server vendor in terms of total performance on the list, but due in large part to the K Computer, Fujitsu is now No. 2, ahead of Cray at No. 3.
Among systems vendors, IBM accounted for 223 systems on the list-or 44.6 percent-while HP had 140 (28 percent).