Blue Pacific adds power to Energy supercomputing

Blue Pacific adds power to Energy supercomputing

The IBM supercomputer, with a theoretical peak of 3.88 TFLOPS, will run nuclear-test simulations

By Patricia Daukantas
GCN Staff

When the newest list of the world's 500 fastest supercomputers comes out later this week, a system called Blue Pacific is likely to join its Energy Department cousins near the top.

The 5,856-processor, $94 million Blue Pacific, built by IBM Corp. to serve Energy's Accelerated Strategic Computing Initiative (ASCI) at Lawrence Livermore National Laboratory, theoretically performs 3.88 trillion floating-point operations per second, or 3.88 TFLOPS, on a single problem.

Under the ASCI program, which aims to maintain the U.S. nuclear weapons stockpile without actual testing, IBM is building a second supercomputer, also destined for the Livermore lab, with 21/2 times the horsepower of Blue Pacific.

The 9,472-processor Intel ASCI Red machine at Sandia National Laboratories led the June 1999 list, compiled by researchers at the University of Tennessee and Mannheim University in Germany. Second on the June list was its ASCI cousin, SGI's 6,144-processor Blue Mountain at Los Alamos National Laboratory. ASCI Red also has a theoretical peak speed of roughly 4 TFLOPS.

The June top-500 list ranked a partial installation of ASCI Blue Pacific at No. 11 with 1,344 processors then up and running.

Blue Pacific is a greatly expanded version of IBM's RS/6000 SP Scalable PowerParallel supercomputer with 32-bit, 332-MHz IBM PowerPC 604e processors, said David Gelardi, IBM's director of RS/6000 benchmarking and applications performance.

Wide Pacific

Everything about Blue Pacific is huge. Livermore's supermachine covers 8,000 square feet; a regulation basketball court occupies 4,700 square feet. The two-thirds of the computer reserved for classified calculations has 2.6T of memory and almost 80T of disk storage, much of which serves as scratch space rather than permanent storage, Gelardi said.

'Within a matter of a few microseconds to a second, we're producing the volume of information that is in the Library of Congress,' said Gilbert G. Weigand, DOE's deputy assistant secretary for research, development and simulation.

Each ASCI supercomputer was designed to produce, in a week or two, simulations that would have taken years to calculate on vector supercomputers, Weigand said.

Despite Blue Pacific's high theoretical peak, Weigand said that on a real problem supercomputers typically sustain only 20 percent to 30 percent of theoretical performance. The peak ratings gauge performance on standardized mathematical problems.

Weigand noted that ASCI goes well beyond the race to produce gigantic hardware. The initiative also funds development of complex codes to model the aging of nuclear weapons and techniques for handling the massive data produced by each simulation.

IBM also is building the next milestone in Energy's computing road map, a 10-TFLOPS machine code-named ASCI White. Delivery to Livermore is scheduled within a year, Weigand said.

ASCI White, now half-completed at the IBM plant in Poughkeepsie, N.Y., will have a 64-bit architecture and a new switching technology, code-named Colony, that is four times as fast as Blue Pacific's interconnects, IBM's Gelardi said.

inside gcn

  • urban air mobility (NASA)

    NASA seeks partners for urban air mobility challenge

Reader Comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above

More from 1105 Public Sector Media Group