U.S. Plans To Overtake China's Supercomputer Top SpotU.S. Plans To Overtake China's Supercomputer Top Spot

The federal government is building supercomputers that will far outstrip a Chinese supercomputer that is now the world's fastest, and has long-term plans for exascale computing, computing at about 400 times the horsepower than any current supercomputer.

China's Tianhe-1A, unveiled last month, set a record of 2.57 petaflops in the latest edition of the quarterly Top500 list of the world's most powerful supercomputers. Another Chinese supercomputer, Nebulae, ranks third on the list.

However, the Department of Energy's Lawrence Livermore National Laboratory and Oak Ridge National Laboratory are both planning to launch computers in 2012 that will be almost eight times faster than Tianhe-1A at 20 pflops, and a federal-state partnership, the National Center for Supercomputing Applications, is also building a supercomputer that may best Tianhe-1A.

The most powerful of the supercomputers currently being built are Oak Ridge's effort, about which little is known other than its existence, and Lawrence Livermore's Sequoia. Sequoia, based on a future iteration of IBM's BlueGene supercomputing technology, is scheduled to be delivered next year and will be operational by 2012. Lawrence Livermore plans to use Sequoia -- which will include 1.6 pbytes of memory and 1.6 million cores -- to do advanced "uncertainty quantification" simulations and weapons science.

The National Center for Supercomputing Applications, meanwhile, is working in a National Science Foundation-funded joint effort with the University of Illinois at Urbana-Champaign (headquarters of the NCSA), IBM, and the Great Lakes Consortium for Petascale Computation to build a computer called Blue Waters. Blue Waters will be used for a wide variety of science research, including biology, astronomy, materials science, engineering, physics, and meteorology. The supercomputer will achieve sustained performance of about 1 pflop and peak performance of around 10 pflops.

The Defense Advanced Research Projects Agency's Omnipresent High Performance Computing program, meanwhile, is developing hardware and software designs for exascale computers and the necessary power management, file systems, and resource management technologies. An exaflop is 1000 petaflops, or a quadrillion operations per second.

The Department of Energy is also doing research to assess the needs and feasibility of exascale computers. For example, the agency's office of science last month announced four Sandia National Laboratory researchers will receive a total of $2.6 million for computational research on technologies necessary to build exascale computers. An IDC research report recently predicted that the agency would ask Congress for $5 billion to develop multiple exascale computers.