A Sleek, Superpowered Machine

By Philip Elmer-DeWitt

With its black frame, red Naugahyde base and transparent plastic panels, it looks like a cross between a recreation-room bar and an aquarium. Its blue- tinted towers, washed by 200 gallons of liquid coolant, bubble and shimmer / like over-heated Lava Lites. Its nickname is "Bubbles," and it bears little resemblance to the computers that most Americans have seen. But the $17.6 million Cray-2 is a computer -- a supercomputer at that -- and it is the fastest one in operation today.

Last week in a brightly lit room at Lawrence Livermore National Laboratory in Livermore, Calif., the first production model of the Cray-2 gurgled and glowed, and a nearby printer spewed out a string of characters: s905. B D/U WO/F 06/04 15:24:22 16a. Software Manager Dieter Fuss stared at the message and interpreted it for the assembled Livermore technicians and executives: "It just came alive and said, 'I'm ready.' "

In that moment, a new era of high-speed computing began. The Cray-2 has the world's largest internal memory capacity (2 billion bytes) and a top speed of 1.2 billion FLOPS (floating point, or arithmetical, operations per second), six to twelve times faster than its predecessor, the Cray-1, and 40,000 to 50,000 times faster than a personal computer. It outdistances the world's half-dozen other supercomputers -- machines specially designed to carry out vast numbers of repetitive calculations at incredible speeds -- and is expected to make short work of problems that have vexed scientists and engineers for decades. Says Robert Borchers, Lawrence Livermore's associate director for computations: "What took a year in 1952 we can now do in a second."

Who needs such blinding speed? At the Livermore Lab and NASA's Ames Research Center in Mountain View, Calif. (which is scheduled to receive the second Cray-2 in September), the new machines will be used for such projects as studying the intense magnetic fields needed for fusion reactors and designing heat shields for future space probes. Meanwhile, demands for ever faster computers are coming from researchers in nearly every discipline of science and engineering, from astrophysics to automobile design. Using more powerful supercomputers, aircraft manufacturers will be able to simulate the airflow around an entire airplane, simplifying their design task. Similarly, energy companies will be able to find more oil and meteorologists to make more accurate forecasts, eventually providing earlier warnings of impending storms like last month's killer tornadoes.

U.S. intelligence agencies depend on supercomputers to sort through the enormous quantities of surveillance data beamed home by ground-based listening posts and orbiting spy satellites. By using supercomputers to simulate explosions, nuclear weapons experts require fewer test explosions to validate their designs. Machines like the Cray-2 are essential to any Star Wars defensive system for locating and intercepting incoming missiles before they re-enter the atmosphere.

Some 150 supercomputers are in use around the world today; most of them were made in the U.S., and more than 100 of them were designed and built by Cray Research of Minneapolis. The new machine, like the Cray-1 before it, was the brainchild of the brilliant and reclusive Seymour Cray, 59, who founded the company in 1972. By densely packing 240,000 computer chips into a C-shaped cabinet 53 in. across and 45 in. high, Cray was able to minimize the crucial limiting factor in supercomputing: the time it takes electric currents to travel from one part of the machine to another.

But by squeezing the chips into a package the size of a hot tub, Cray created another problem. Unless it could be dissipated, the heat generated by electrons flowing through the tightly packed circuit boards would quickly melt the machine. Cray's answer: flooding the circuits with a continuous flow of Fluorinert, a liquid coolant that, incongruously, is also used as an artificial blood plasma. "It's a brilliantly innovative solution," says Astrophysicist Larry Smarr, director of a new supercomputer center at the University of Illinois at Urbana-Champaign. "It's typical of Seymour's genius."

In the past few years, several Japanese firms have announced machines they claimed were faster than the U.S.'s most powerful supercomputers. But Cray Research, with two competing design teams, continues to hold the lead. Last month the company announced a new customer for its hot X-MP supercomputer line: Nissan Motors. Boasts Cray Chairman John Rollwagen: "We compete with the Japanese even on their home ground."

At Cray laboratories in Chippewa Falls, Wis., Designer Steve Chen is hard at work on the X-MP's successor, while Seymour Cray is devoting his time to the Cray-3. Due out in 1988, the Cray-3 will have an awesome 8 billion-byte memory. "As far back as I can remember, I was very proud to make factor-of- four improvements from one generation of machines to another," Cray said at a rare public appearance last month. "Instead, we're moving forward by factors of ten."

The next generation of supercomputers, by most accounts, will achieve its gains not by using faster processors but by using more of them. The Cray-1 had one main processor. The Cray-2 has four, and the Cray-3 may have as many as 16, all working in parallel. University researchers are now attempting to yoke dozens, even hundreds, of microprocessors together. "Multiprocessing is the biggest development in 40 years," says Illinois' Smarr. "We're entering what will be looked back on as the birth of new kinds of computers." One such machine under development at New York University has already been given a new name: the ultracomputer.

With reporting by J. Madeleine Nash/Chicago and Dick Thompson/San Francisco