March 25, 1996
U.S. Evaluates Fast Computer From Japan
By JOHN MARKOFF
Japanese computer maker may be on the verge of achieving what had previously been unthinkable in this country: selling a supercomputer to a U.S. government laboratory.
Two of the three finalists now bidding for a $13 million to $35 million supercomputer being sought by the Colorado-based National Center for Atmospheric Research, or NCAR, are Japanese electronics giants: Fujitsu Ltd. and the NEC Corp.
According to several people familiar with the contract review, Fujitsu and NEC are seriously challenging the U.S. leader in the field, Cray Research Inc., which was recently acquired by Silicon Graphics Inc.
The impending announcement of a winner has piqued interest by both the U.S. and Japanese governments, because supercomputers have long been symbols of national technical prowess and have remained at the heart of bitter trade disputes between the two nations.
In Japan, there are more than 100 U.S.-designed supercomputers in government-funded laboratories. But a U.S. federal agency has never turned to Japan for a supercomputer and NCAR has seldom strayed from a Cray -- of which it now has five. So NCAR's award of its contract to a Japanese computer maker could establish a beachhead in territory that for three decades has been the unassailable property of American computer makers.
The center's director, Robert Serafin, said that precedents and trade politics would take a back seat to computing performance as the evaluation proceeded. "We have undertaken the equivalent of an international procurement," said Serafin, who declined to handicap the contestants. "We're playing it straight as far as the technical and scientific factors are concerned."
The laboratory is preparing to buy a new computer to do advanced global climate modeling into the next century.
The award, which will be made next month or in early May, will come after an exacting yearlong review in which NCAR scientists have put dozens of supercomputers through painstaking performance tests. Because of the political sensitivity of the contract, NCAR has been unusually diligent in its testing procedures, industry experts said.
Cray has long dominated the market for weather forecasting supercomputers, but its status was shaken last year after the European Center for Medium-Range Weather Forecasting in Berkshire, England, purchased a $40 million system from Fujitsu. The European market is believed by many industry executives to be less biased by national political issues and a better gauge of pure technical factors.
The supercomputing business of the United States' national laboratories like NCAR has been the sole province of American companies -- beginning in the mid-60s with the Control Data Corp. and then Cray Research, and more recently with start-up companies like Thinking Machines and Maspar.
And trade politics have long played a role. In 1987, for example, the U.S. government successfully pressured the Massachusetts Institute of Technology not to buy a computer from NEC.
In 1990, Japan and the United States signed a trade agreement declaring that each nation would open its supercomputer market to the other. But even after that, in September 1990, an indication by the Lawrence Livermore Laboratory that it might evaluate a Japanese machine drew such criticism from Congress that lab officials backed down.
Since then, the inability of Japanese computer makers, who now build many of the world's fastest supercomputers, to sell even a single supercomputer to a U.S. research laboratory or university -- despite building many of the world's fastest machines -- has remained a sore point for the Japanese. And they, in turn, have routinely been accused by U.S. corporate executives and government executives of maintaining their own market restrictions.
Now, the impending NCAR supercomputer award comes at a time when high-technology trade frictions between the United States and Japan are again heating up. Cray Research recently, and quietly, urged the Clinton administration to put new pressure on the Japanese in the wake of the company's two losing bids in the Japanese government market, according to the newsletter Inside U.S. Trade. The publication reported that of 18 contracts for Japanese supercomputers in the last year, only four were awarded to U.S. companies.
"I have to say it feels like it has become tougher to do business in the public institutions in Japan," said Robert Ewald, president and chief operating officer of Cray Research.
But both Ewald and his Japanese competitors said they saw no evidence that political issues have played a role in the decision making process at NCAR up to this point.
"We're in the hunt," said Tom Miller, sales and marketing vice president for Fujitsu's American subsidiary. "We feel that if this in fact turns out to be a truly open procurement, it's a tremendous opportunity for us."
Despite trailing in the market for the world's fastest computers, the Japanese have inexorably closed in on the speed advantage of American companies like Cray, IBM, and Intel. Today, the world's fastest supercomputer, according to a measuring system maintained by Jack Dongarra at the University of Tennessee, is a supercomputer made by Fujitsu that was installed at the National Aerospace Laboratory in Japan in 1993.
Moreover, nine of the top 20 spots on Dongarra's list of the world's 500 fastest computers are now held by either NEC or Fujitsu. Currently, massively parallel supercomputers made by the Intel Corp. occupy the No.2 and No.3 positions on the list.
Fujitsu is now the world leader in absolute speed because the company has designed the world's fastest single processor, capable of 2.2 billion mathematical calculations a second.
Measuring computer speed is a tricky proposition, however. Dongarra's numbers measure pure calculating speed, akin to measuring the performance of a dragster, designed to go down a race-car track in a straight line as quickly as possible. Some computer experts argue that true performance is better measured by a balanced set of factors, more like a Formula One racing car that can also go through turns quickly.
That would mean that despite trailing on pure speed, Cray may still be able to win the contest based on its software and other factors.
"I think that NCAR could safely go with Cray on the basis of the numbers alone," said C. Gordon Bell, a veteran computer designer.
Copyright 1996 The New York Times Company