The computer that defeated world chess champion Garry Kasparov has just been upstaged by IBM's latest supercomputer with an estimated peak performance of 12.3 trillion operations per second, or 1,000 times more powerful than the Deep Blue chess machine.
The new supercomputer performs at 12.3 teraflops, a figure that sets a new benchmark for the workhorses used by government agencies and universities in scientific problem solving and by commercial entities that perform number crunching on a massive scale.
IBM Thursday announced it has begun shipping its latest RS/6000 SP supercomputer - on 28 tractor trailer rigs - from its facility in Poughkeepsie, New York, to the Department of Energy's Lawrence Livermore National Laboratory in Livermore, Calif. The Energy Department will use the computer, which it refers to as ASCI White, short for the Accelerated Strategic Computing Initiative White Partnership, to simulate nuclear explosions and monitor the U.S. nuclear stockpile.
At 12.3 teraflops, ASCI White is the first computer to exceed the double-digit teraflop speed barrier and it earns the title of the world's fastest supercomputer with plenty of breathing room. ASCI White is 512 computers linked together that take up two basketball courts and is the successor to another RS/6000 SP currently in use at the Livermore lab that can perform about 3 teraflops per second.
Though the 12.3-teraflop benchmark is impressive, the measurement is the supercomputer's peak performance with every microprocessor doing all it can all the time, said Earl Joseph, research director of high performance systems at market research company IDC. More impressive is the machine's ability to perform a sustained 3 teraflops while computing an actual problem, Joseph said. Until now, the record was a sustained 1.2 teraflops.
IBM's supercomputer business has picked up momentum over the past 18 months, last year becoming the market leader with an estimated 30% share of the worldwide supercomputer market, which was valued at $934 million last year, according to IDC. The Energy Department is among 8,000 IBM supercomputer customers, most of which are commercial entities.
The department is paying $110 million for ASCI White, but the Energy Department is getting more than it requested in its original order. IBM won the contract in 1996, and 2 years later the department exercised an option included in the contract that allowed the Energy Department to ask for a system with 10 teraflops, said Paul Messina, acting director of the ASCI office in the Energy Department's National Nuclear Security Administration.
IBM managed to exceed the Energy Department's teraflop requirement through a combination of faster copper processors, a new switch and sophisticated software, said Jim Jardine, project manager for ASCI White.
The system's 8,192 processors use copper wiring technology patented by IBM in 1997 that lets them run cooler than aluminum processors and provides better conductivity. The new switch connects to each node three times faster than the switch IBM currently sells in its supercomputers, Jardine said. In addition, the software developed especially for the system with help from Livermore software developers coordinates the activity on each computer and the switch so that the system runs as if all its parts were one.
"There are two big components of the technology. One is that each (of the 512) computers can do a lot of work," Jardine said. "The other component, which isn't quite so obvious, is to turn the 512 computers into one big computer."
Although the switch in ASCI White is a type that has been around for a while, the technology has been steadily improved, and the newest version seems to have "gotten over the hump," making it the most impressive part of the latest supercomputer, Joseph said.
"If they didn't have the special switch technology and software that hooks all these computers together, it would be the same as buying 512 separate computers," Joseph said. "A lot of companies have gotten eight or 16 nodes working together, but to have 512 working together, that's significant."
The supercomputer's job once it's operational in the fourth quarter of this year will be to simulate nuclear explosions, a challenge for physicists because of the tremendous range of temperatures, pressures, speeds and time scales that occur, Messina said. Even small changes can make a big difference to the outcome.
Simulations also have become necessary because of a shift in U.S. policy away from actual tests on islands in the Pacific or in the desert.
The Energy Department also will use the new supercomputer to monitor the condition of the U.S. nuclear stockpile by modeling the materials as they age. The modeling helps scientists keep check on what happens to the bombs and other materials as they age. For example, leakage, and the condition of the casing of the bomb and its electronics after exposure to radiation are concerns scientists have to investigate.
Messina said there's tremendous demand for working through problems on the Energy Department's existing IBM supercomputer at the Livermore lab and on two other Energy Department supercomputers - a Silicon Graphics Inc. supercomputer at Los Alamos National Laboratory in New Mexico and an Intel supercomputer in use at Sandia National Laboratories.
"Right now the 3-teraflop machines are so busy with our problems that there is contention and we have to schedule time," Messina said. "It's amazing, but the problems are really there. We can't wait for [ASCI] White to come into production."
IBM, in Armonk, N.Y., is at www.ibm.com/.
RELATED LINKS
