Computer Chip off the Old Block

By Philip Elmer-DeWitt

For nearly two decades, the name Cray Research has been synonymous with supercomputers, those lightning-fast machines used for everything from locating oil deposits to designing nuclear warheads. Not only had Cray seized nearly two-thirds of the world market for number crunchers in the $5 million- to-$25 million range, but it held exclusive license to sell any machine made by Seymour Cray, who is to supercomputers what Alexander Graham Bell was to the telephone.

Now Cray and the company he founded have decided to go their separate ways. In an unexpected move, the firm announced last week that it was splitting into two rival entities: Cray Research, based in Minneapolis, and Cray Computer, based in Colorado Springs and headed by Seymour Cray. The new company, financed with $150 million in cash and equipment from its parent firm, will devote itself to developing the long-awaited Cray-3, a computer that will compete head on with the next generation of supermachines produced by Cray Research. "It's a stunning development," says Gary Smaby, an analyst with , Needham & Co. "For a company to set up and fund a direct competitor must be unprecedented."

The dramatic breakup was the latest of several surprises that rocked the intensely competitive industry this spring. In April NEC, one of Japan's three supercomputer makers, announced a machine it claims is eight times faster than the speediest Cray. A week later Cray's crosstown rival Control Data declared that after five years and $238 million in losses, it was closing its supercomputer subsidiary, ETA Systems. That left Cray as the last U.S. company still racing the Japanese for pre-eminence in what both countries view as a technology critical to the future of science and industry.

All this made the U.S. supercomputer effort even more dependent on one man: Seymour Cray. At 63, Cray is one of the most enigmatic figures in computer science. A restless, rugged individualist of legendary idiosyncrasy (for many years he made a point of building a new sailboat every winter and, inexplicably, burning it in the fall), he has devoted his professional life, first at Control Data and later with his own firm, to building the world's most powerful computers. His track record: an unequaled series of five major computer designs dating back to 1960, each for what would be the fastest machine of its time.

The Cray-3 was to be his most impressive to date. People who have seen prototypes describe it as a technological tour de force. To minimize the distance electrons have to move within its components, Cray is squeezing chips capable of 16 billion calculations per second into a tight octagonal package 32 in. across, the size of a small coffee table. The computer's basic building blocks are 4-in. by 4-in. modules each bejeweled with 1,024 chips and threaded with more than a million interconnects of braided gold wire thinner than a hair.

But designing a supercomputer and getting it to market are two different things, and with his latest machine Cray may have pushed the technology one step too far. Not only does the 16-processor Cray-3 contain four times as many central calculating units as the Cray-2 (an increase that more than quadruples its complexity), but it relies on an as-yet-unproved technological advance: replacing silicon chips with faster ones made of gallium arsenide. Add to Cray's headaches the fact that his new computer is so compact that assembly by hand is difficult. Before production could begin, he would have to endow robots with the manipulative skills of a jeweler or watchmaker.

^ Cray Research, meanwhile, has had other troubles. Sales are sluggish, profits are down and its stock price has plummeted. With R. and D. expenses growing nearly 35% a year, Chairman John Rollwagen found himself having to choose between two projects: the Cray-3 and the C-90, an extension of the company's bread-and-butter Cray Y-MP line.

As Rollwagen tells it, the decision turned on a chat he and Cray had four weeks ago in Colorado Springs. "I said to him, 'It's not working, is it, Seymour? It isn't feeling right.' " The two discussed options short of a total split, but Cray kept pressing. "It's almost like he forced me to turn the page," says Rollwagen. "He said, 'Isn't there ((an option)) that would be even cleaner? Let's get on to that one.' It just became very clear to the two of us that this was the right thing to do."

Reaction was swift. On Wall Street Cray's stock fell 10% in one day. In Japan some thought they smelled a "political maneuver." Since U.S. agencies like to have at least two bidders on any contract, the exit of ETA opened a window of opportunity for Cray's Japanese rivals. The Cray split, they suspect, may have been designed to close that window. Cray officials do not deny it. Chuckles one: "They got the message in a hurry."

Surprisingly, given the relative sizes of the two Crays, some experts voice more concern about the future of Cray Research than they do about Cray Computer. Few doubt that the smaller spin-off firm will be able to raise all the money it needs. As John Sell, president of the Minnesota Supercomputer Center, puts it, "Seymour is magic in this business." Whether Cray Research can flourish without its founding genius remains to be seen. Analysts say that within three to five years it should be clear whether the company has wisely cut its losses or created a killer competitor by trading away its most valuable asset.

CHART: NOT AVAILABLE

CREDIT: TIME Chart by Joe Lertola

CAPTION: A HOUSE DIVIDED

With reporting by Marc Hequet/Minneapolis and J. Madeleine Nash/San Francisco