At the International Supercomputing Conference 2011 (ISC '11) a new Top 500 list of supercomputers was unveiled on Monday. Based on the new rating, Fujitsu K supercomputer with 8.162PFLOPS (quadrillion floating-point operations per second) performance is the world's highest-performing supercomputer.
The TOP500-ranked K computer system, currently in the configuration stage, has 672 computer racks equipped with a current total of 68 544 microprocessors. This half-build system achieved the world's best LINPACK benchmark performance of 8.162 petaFLOPS (quadrillion floating-point operations per second), to place it at the head of the TOP500 list. In addition, the system has recorded high standards with a computing efficiency ratio of 93.0%. This is the first time since June 2004 that the Japanese supercomputer "Earth Simulator" has been ranked first on the Top 500 list.
"I am delighted that we were able to achieve this result, made possible through the tremendous efforts of all involved, despite the impact of the Great East Japan Earthquake. In particular, I am sincerely grateful to our partners in the Tohoku region for their commitment to delivering a steady supply of components, even though they themselves were affected by the disaster," said Michiyoshi Mazuka, chairman and representative director of Fujitsu Limited.
Fujitsu K supercomputing system will be comprised of more than 800 computer racks, each installed with Fujitsu's SPARC64 VIIIfx central processing units (CPUs). Each of these eight-core processors possesses a computational performance of 128 gigaFLOPS, and has a degree of reliability inherited from Fujitsu's mainframe technology. The CPUs are also highly energy efficient, with a world-class processing power of 2.2 gigaFLOPS per watt.
The supercomputer will comprise of over 80 000 of these processors in an interconnected network, utilizing the world's first six-dimensional mesh-torus topology developed by Fujitsu. This will permit the system to be used more efficiently, as multiple processes can be flexibly allocated to groups of processors.
When configuration of the K computer is complete in 2012, it is designed to achieve LINPACK performance of 10PFLOPS. It will be widely used in a variety of computational science fields where it is expected to contribute to the generation of world-class research results. The K computer is a wholly made-in-Japan supercomputer, from the research and development of the processors, to system design and manufacturing.