IBM's ASCI White isn't for playing chess. It crunches numbers for scientists, faster than any other computer in the world, according to rankings released Friday by the TOP500 list of supercomputers.
While another IBM supercomputer, Deep Blue, earned headlines for beating Garry Kasparov in 1997, ASCI White works on classified nuclear weapons research problems for the Lawrence Livermore National Laboratory in California.
IBM has 215 of the top 500 fastest supercomputers and five of the top 10, according to the list. Sun Microsystems Inc. placed second with 92 supercomputers, while Silicon Graphics Inc. had 67 systems on the list for third. Almost half of the systems are used in commercial applications, not scientific applications.
"The commercial market is getting much, much bigger," said Irving Wladawsky-Berger, IBM's vice president of technology and strategy. "The price of the machines keeps going down, so that makes them more attractive for commercial applications. And data mining is exploding in the market place. Because of the Web, we are accumulating information at an exponential rate."
For instance, companies like Charles Schwab & Co. use an IBM supercomputer as a giant Web server for their online clients, he said, estimating that the market for supercomputers is US$5 billion a year and growing. The convergence of computer science and biotechnology for genome-related projects could particularly fuel growth, Wladawsky-Berger said.
The semi-annual TOP500 list is compiled by supercomputer researchers. The list's release precedes the weeklong SC 2000 supercomputing conference in Dallas, which begins Saturday. The compilers use a benchmarking standard called the Linpack test to measure speed and computing power. Using the Linpack yardstick, ASCI White computes at 4.9 teraflops per second.
Overall supercomputing power grew rapidly, according to the TOP500 list maintainers. Since June, 231 systems dropped off the TOP500, the second highest replacement rate ever, according to the list's Web site. The total accumulated performance of the fastest 500 systems is 88.1 teraflops per second, an 61.5 percent increase in power over the last six months. The slowest computer on the TOP500 list, an IBM's SP PC604e, runs at 55.1 gigaflops per second.
The type of computing has changed as well. Just 11 of the TOP500 are pure SMPs (shared memory processors), down from 121 six months ago. Of the top computers, 112 are clustered SMPs. That's due to a change in the kind of computing now being done, Wladawsky-Berger said.
"Supercomputing is best known for scientific applications ... engineering simulations, or modeling weather patterns for global warming for example," he said. "More and more, people are solving larger problems that require more processing power, and as the size of supercomputing applications has grown, you reach a certain limit and you can't go up" using pure SMP architecture.
A more parallel architecture allows researchers to add processors without theoretical limit. The bottleneck becomes the software.
"Can you decompose the problem so that lots of processors can work on it at once," he said. "A huge part is the software and the kind of problems to solve."
The TOP500 list can be viewed at http://www.top500.org/. IBM, in Armonk, New York, can be reached online at http://www.ibm.com/.