For the first time, China has placed in the list of top supercomputers. A U.S.-based supercomputer is still first in the TOP500 list, but two Chinese machines are now in the top 10.
Oak Ridge National Laboratory's Cray XT, running at 1.75 petaFLOPS, is in first place. Known as the Jaguar, it uses an AMD x86_64 Six Core, and it can perform calculations in one second that would take a PC 10 hours. A petaFLOP is 1,000 trillion calculations per second. On the drawing boards at various labs are exascale computers, which could do one quintillion calculations a second -- that's one million trillion.
Top Speed of 2.98 PetaFLOPS
The top Chinese supercomputer, called Nebulae, runs at 1.2 petaFLOPS, about twice as fast as the previous fastest computer from that country. Theoretically, Nebulae can run as fast as 2.98 petaFLOPS, which, if accomplished, would vault it into first place. It is built on a Dawning TC3600 Blade system, using Intel X5650 processors and Nvidia Tesla C2050 GPUs.
According to news reports, a faster machine is currently being built at China's National Supercomputer Center in Tianjin. The U.S. has one other petaFLOP supercomputer, which was replaced in first place by Jaguar. However, the TOP500 only lists publicly available machines, and doesn't include the top-secret supercomputers in government use, such as those at the U.S. National Security Agency.
The next supercomputers in the top 10, in order, are IBM's Roadrunner, the National Institute for Computational Sciences Kraken XT5 at the University of Tennessee, IBM's Jugene, the NASA Ames Research Center's Pleiades, China's National SuperComputer Center's Tianhe-1, IBM's BlueGene/L, Argonne National Laboratory's Intrepid, and the Sandia National Laboratories Red Sky.
Supercomputers are used for processing-intensive tasks, such as designing airplanes, exploring for oil, virtual testing of vehicles in wind tunnels, processing telescope data, or predicting weather.
Began in 1993
The TOP500 list presents the 500 most powerful computer systems that are commercially available. The project began in 1993 to track trends in high-performance computing, and comes out twice yearly. It's managed by representatives from the Lawrence Berkeley National Laboratory, the University of Mannheim in Germany, and the University of Tennessee.
The Linpack benchmark is used as the comparative yardstick, since "fastest" can depend on the specific application being run. Linpack was chosen because the benchmark is available for virtually all supercomputers.
The benchmark requires the system to solve a series of linear equations, with the user allowed to scale the problem size and to optimize the software to provide the best performance. It is not used to indicate overall system performance, given the limitations of a single benchmark, but the TOP500 group said the benchmark can be said to reflect "the performance of a dedicated system for solving a dense system of linear equations."