And Then There Was One

By MICHAEL QUINN

"Monday was a sad day in American computing," declared Doyle Knight, president of New Jersey's John von Neumann National Supercomputer Center. What dismayed Knight was the announcement last week by Control Data, based in Minneapolis, that it would halt its money-losing six-year foray into the growing world market for supercomputers. The decision leaves Cray Research, its crosstown rival and the industry leader, as the only U.S. supercomputer maker at a time when Japan's industry is moving vigorously into the field.

First introduced by Cray in 1976, supercomputers possess vast number- crunching power, which has become increasingly valued in applications ranging from oil exploration to the design of new drugs. Supercomputer sales are expected to reach $1.2 billion this year on shipments of 130 to 150 machines, up from $925 million in 1988. Control Data had sold 34 of its ETA- model supercomputers, or about 12% of the world's installed machines (vs. 63% for Cray).

The ETA machines offered such innovations as ultradense clusters of circuit boards cooled by liquid nitrogen, but lacked adequate software. Control Data lost about $200 million on supercomputers during the past three years, and Chairman Robert Price thinks the venture would have consumed plenty more. Said he: "Let's put it this way: there are less risky ways to bet $200 million in the computer industry."

Japan's quest to build the high-powered machines began in 1981, when the government arranged a $136 million supercomputer-research project. Three Japanese manufacturers -- Fujitsu, NEC and Hitachi -- account for 24% of the supercomputers sold to date. So far only U.S. and Japanese companies have entered the race. While Cray's machines still lead in worldwide sales, Japanese manufacturers may be pulling ahead by some measures of supercomputer performance, notably processing speed. Earlier this month NEC introduced a new series, called SX-3, billed as the world's fastest supercomputers, even though the machines will not be available until June 1990. Cray discounts NEC's claim to top speed, contending that such measurements are based on "theoretical peak performance" figures rather than a practical application. NEC insists that the SX-3's key elements have been tested.

The exit of Control Data could aggravate U.S.-Japan trade friction over supercomputers. Says Etsuro Yamada, a spokesman for Fujitsu: "The fact is that Control Data lost in a fight with Cray, but that may not be the way the Americans will look at it." The U.S. has long complained about the Japanese government's failure to buy U.S.-made supercomputers. The two countries signed a 1987 accord in which Tokyo agreed to eliminate discrimination against U.S. supercomputer makers in the purchasing procedures of Japanese government agencies and universities. But since then, Tokyo has failed to buy a single U.S. supercomputer.

The bright spot for the U.S. is that several companies, large and small, hope to enter the field. Among the contenders is IBM, which in late 1987 formed a venture with former Cray designer Steve Chen to develop a line of advanced supercomputers. Allan Weis, a vice president in IBM's Data Systems division, asserts, "We're very serious about the supercomputer market. The Japanese are formidable competitors, but IBM and Cray are very formidable too." They had better be, or the supercomputer could go the way of the videocassette recorder.

With reporting by Marc Hequet/St. Paul and Thomas McCarroll/New York