The US Department of Energy Oak Ridge National Laboratory's new Titan system, a Cray XK7, has been anointed as the world's fastest supercomputer in the newly released 40th edition of the Top500 compilation of the world's fastest supercomputers.
Titan knocked from the top spot the DOE Lawrence Livermore National Laboratory's Sequoia supercomputer, an IBM BlueGene/Q system that topped the last Top500, compiled in June.
Under the Linpack benchmark, Titan executed 17.59 petaflops (quadrillions of calculations per second). By contrast, Sequoia executed 16.32 petaflops during this round.
Cray's machine runs 560,640 processors, including almost 300,000 AMD Opteron 6200 series cores and over 261,000 Nvidia K20x accelerator cores.
Also in the top five were the RIKEN Advanced Institute for Computational Science's K supercomputer (a Fujitsu system performing at 10.5 petaflops); The DOE Argonne National Laboratory's Mira, a BlueGene/Q system (8.16 petaflop/s); and the German Forschungszentrum Juelich's Juqueen, also an IBM BlueGene/Q system (4.14 petaflops).
This edition of the Top500 marks the 20th anniversary of the list. Researchers from the University of Mannheim, Germany; Lawrence Berkeley National Laboratory; and the University of Tennessee, Knoxville, compile the Top500 twice a year, in June and November. Inclusion is voluntary, and the Linpack tests are run by the managers of the participating systems.
The keepers of the list will further discuss the results of this edition at the SC12 supercomputing conference, being held this week in Salt Lake City.
As it has since its launch, the Top500 continues to show the rapid pace at which supercomputers are developed. This iteration has 23 systems demonstrating petaflop performance, just four and a half years after the National Center for Supercomputing Applications' Cray-based Roadrunner, the first Petaflop-scaled system, debuted on the list.
The floor of the Top500 - the lowest charting computer on the list - has moved up to 241.3 teraflops from 172.7 teraflops six months ago.
The compilation reveals other trends in the supercomputing arena as well. This edition has 62 systems using accelerator and co-processor technology such as Nvidia GPUs (graphics processing units), up from 58 six months ago. For internal interconnects, InfiniBand technology gains favour - being used on 226 systems, up from 209 systems six months ago - while Gigabit Ethernet usage has shrunk, 188 systems, down from 207 six months ago. And systems continue to use multicore processors - 84.6% of the systems on the list use processors with six or more cores
Geographically speaking, the US continues to dominate the list, hosting 251 of the top 500 systems. Europe has 105 systems and Asia has 123 supercomputers on the list. Among countries, China has the second largest number of computers, with 72 on this list.