NASA Supercomputer Looks to Blast Off

NASA is preparing to rocket its research supercomputer into the petaflop era.

The space agency has signed an agreement with Intel and SGI to update its supercomputer at the Ames Research Center in Moffett Field, Calif. NASA hopes the project, dubbed “Pleiades,” will produce a supercomputer capable of breaking the petaflop mark—1 quadrillion calculations per second—by 2009.

By 2012, SGI, Intel and NASA plan to increase the performance to 10 petaflops.

The news that NASA and its partners are planning to build a machine that will first break and then move past the petaflop mark has become a common refrain in the supercomputer industry within just the last year. Both Sun Microsystems and IBM have each announced plans to build new supercomputers that pass the petaflop mark. Cray, one of the world’s best-known supercomputer companies, is also working toward that goal.

It’s possible the first petaflop machines could enter the Top 500 supercomputer list in June, when the organization that ranks these machines updates the list for the first time since November. IBM and its BlueGene/L system at the Department of Energy’s Lawrence Livermore National Laboratory currently hold the top spot with a maximum performance of 478.2 teraflops.

Right now, the NASA supercomputer at Ames, called Columbia, ranks 20^th on the list, with a performance of 51.9 teraflops. The new system, with the improvements that Intel and SGI are planning, should offer 16 times more computing power than the current Columbia system.

To show how far the industry has gone since the Columbia system was first installed, the machine originally ranked second on the Top 500 when it debuted in 2004 with the same performance of 51.9 teraflops, or 51.9 trillion calculations per second. Since then, Columbia has tumbled down the list as IBM began pushing the limits with its line of BlueGene systems.

SGI first built the NASA supercomputer in 2004.

That system, which uses more than 10,000 processors, is built on Intel’s Itanium architecture. The May 7 announcement from Intel and SGI did not indicate if the new system would continue to use the Itanium processors or switch to Intel’s x86 processors.

Earlier this month, Intel announced that it had signed an agreement with Cray to use its x86 processors with that company’s supercomputers. The agreement, the first between Intel and Cray, shows the increasing acceptance of x86 processors within the field of high-performance computing (HPC) and Intel’s desire to chase after a market that had been dominated by highly specialized processors or chips that were built with IBM’s Power microarchitecture.

With its next-generation architecture—Nehalem—Intel will introduce a technology called “QuickPath,” a high-speed, chip-to-chip interconnect that should create a low-latency way to connect thousands of processors together inside supercomputers and HPC machines.

The announcement from NASA was not the only supercomputer news this week. On May 6, Dell announced it had built a new supercomputer on the campus of Purdue University. The cluster system, called “Steele,” is made up of 812 PowerEdge 1950 nodes. Each node supports two, quad-core Intel Xeon processors, which should allow for a peak performance of 60 teraflops.

While not as prolific as IBM or Hewlett-Packard, Dell has 24 supercomputers listed within the Top 500, and the 60 teraflop performance of the Purdue machine would place it within the Top 20. For its part, SGI has 22 machines listed on the Top 500, including one that breaks 100 teraflops of performance.