Cray Research challenged by makers of cheaper `mini-supercomputers'

Industry wondering what company will do about it

John Markoff; New York Times News Service
Colorado Springs Gazette Telegraph

September 17, 1989

Cray Research Inc., the leading American maker of the fastest kind of computer, faces a growing challenge from rival machines that are less powerful but also far less costly.

The Minneapolis-based supercomputer maker has already found its top-of-the-line market threatened by Japanese companies and IBM.

The threat at the low end comes from makers of "mini-supercomputers." These machines have as much as a quarter of the performance of a Cray for some applications, but sell for as little as a tenth of the price of the most powerful machines, which can cost $2.5 million to $25 million. The leader in the mini-supercomputer field is Convex Computer Corp.

So much has the competition from below concerned Cray that the company, the world's largest supercomputer maker may respond with a scaled-down machine of its own - a "Crayette" in industry jargon.

Such a move would be fraught with risk for Cray, which prides itself on producing limited quantities of handcrafted systems that outrace the competition.

Entering the mini-supercomputer market would mean lower margins, higher marketing expenses and an investment in costly mass-production technology.

"Its a different ball game and a different class of customer," said Gary P. Smaby of Smaby Group Inc., a consulting firm in Minneapolis.

Supercomputers which can quickly handle billions of computations, are used for mathematically intensive scientific and engineering problems, while general-purpose computers are used more for sorting and filing information.

Progress in everything from aircraft technology to mapping the human genetic code depends in large part on extremely fast machines.

But because high-performance computing technology is advancing so quickly, some potential customers are stopping to consider whether it is wise to invest $20 million in a machine that may quickly grow obsolete.

"It's a big investment that will sit there for five years," said C. Gordon Bell, vice president of research at Ardent Computer Inc., a mini-supercomputer company in Sunnyvale, Calif.

Analysts say the merger of Ardent and Stellar Computer Inc. may provide another strong competitor at the low end of the market.

The mini-supercomputer "requires less overhead, it's cheaper to buy and cheaper to cool, and you can get one for your own department," said Peter Gregory, who formerly headed strategic planning at Cray and is now chairman of Myrias Research Corp., a start-up mini-supercomputer company in Edmonton, Alberta.

"And the real performance gap is narrowing between what you can get out of a $1 million and a $20 million computer."

Both classes of computers are increasingly being used by large corporations for applications as diverse as searching for oil and simulating the impact of automobile collisions.

The energy and automotive industries pioneers the commercial use of supercomputing, but they are being joined by a wider variety of corporations.

For example, Eastman Kodak Co. has used a Cray supercomputer at the National Center for Supercomputer Applications, one of five federally sponsored supercomputer centers, to figure out how to improve the flow of liquid plastics into moldings used to make camera parts.

Cray considered a scaled-down "departmental" supercomputer in the early 1980's. But the project never got off the drawing board and was terminated in 1983, said Gregory, who headed the study.

At recent meeting of analysts, the company confirmed that it was again contemplating a shift in strategy.

A less powerful Cry computer would be air-cooled, in contrast to the company's super-computers that require costly liquid cooling systems, and would sell for less than $1 million. It would serve two functions, computer industry analysts said: defending the low end of Cray's market and "seeding" the market for high-end supercomputers.

Convex is Cray's greatest threat at the low end.

The Dallas-based mini-super-computer maker thinks its air-cooled machines, which cost less than $1 million, will have the same impact on the supercomputer market as work-station vendors have had on the business market for mainframe computers.

"Convex will continue to give Cray fits at the low end," said Steve Milunovich, a computer industry analyst at First Boston Corp.

Convex was founded in 1981. Steven J, Wallach, vice president for technology and chief designer of the company's first computer, the Convex C-1, was leader of the design team for Data General's Eclipse MV minicomputer line.

Wallach is now designing a third generation of the company's computers.

After heated competition and an industry shakeout, two other mini super-computer makers are still in the running: Sequent Computer Systems Inc., based in Beverton Ore., and Multiflow Computer Inc., based in Branford, Conn.

Convex reported record revenue of $105.6 million last year, up 52 percent from a year earlier, while Cray's revenue was $756.3 million, up 10 percent. But Convex has sold more than 500 machines to 325 customers, compared with Cray's 218 machines and 166 customers.

And while Cray's growth flattened in 1988 and 1989, Convex's revenue grew 52 percent last year. As a result, many industry analysts said Cray now thinks it must respond with a less costly machine.

Cray by no means loses all of its battles to lower-priced competitors. The company repurchases its used machines and sells them at deep discounts. It recently defeated Convex by selling an inexpensive used machine to an Arizona university.

The danger for Cray is that new users in the pharmaceutical and transportation industries are deciding on less expensive mini-supercomputers from Convex.

At the same time, Cray's competitors warn that entering the lower-cost, broad-based market may not be immediately profitable or easy.

"Cray has been pre-eminent in their market," said Bob Paluck, chairman and president of Convex.

"If they come out with a million-dollar machine they won't see any profits relative to their main business. But it will cause a complete restructuring of their business to get into this market. They will have to change everything from marketing to development."

Despite the innovative uses of supercomputing, the high end of the market has not been developing as quickly as Cray had hoped. John A. Rollwagen, chairman of Cray research, has criticized large American corporations for not being quicker to adopt the technology to make productivity gains in developing products.

He acknowledges that Cray's problems have been exacerbated by the fact that the U.S. Government, its largest customer, has cut its spending on supercomputers.

Direct sales to government agencies accounted for 44 percent of the company's of the company's 1988 revenue. And while revenue from the government grew 68 percent in 1987 and 13 percent in 1988, analysts forecast that such sales will grow only 10 percent this year and will be flat in 1990.

Cray has had other fundamental problems. Unable to support simultaneously its two $100 million-plus supercomputer development projects, the company this year spun off a division run by its founder, Seymour Cray.

The new company, the Cray Computer Corp. of Colorado Springs, has essentially created new competition for Cray Research at the high end of the concerned that both companies may soon be vulnerable.

Despite widespread industry speculation that Cray has either begun to design its own machine to sell for less than $1 million, or is considering a marketing agreement to resell another manufacturer's computer, Cray officials have said that no decision has been reached.

"The issue of a departmental computer is something that we keep looking at, and there us more and more reason to have some type of product," Rollwagen said.

But he insisted that those who compare supercomputers and mini-supercomputers with mainframes and personal computes were mistaken.

"There's a reason for having these large-scale machines," he said. "They can do things that are simply impossible on less powerful computers." Such brute power can be crucial. Only the most powerful Cray models have enough speed and memory to model advanced supersonic aircraft.

Many of Cray's customers agree. "We're basically not interested unless a machine is capable of supercomputing speeds," said Robert Borchers, associate director for computation at Lawrence Livermore Laboratory. "For running big problems you need big machines."

Indeed, some analysts argue that Cray's greatest danger would be shifting its focus from what it has done best for 17 years: building the world's fastest computers.

"Even with all the trauma that Cray management has gone through, it still feels that its mission is based in the high end of supercomputing," said Jeffry Canin, an independent computer analyst in San Francisco.

Label: BUSINESS

Caption: Associated Press - Many Cray Research customers require a full-size Cray supercomputer such as the Cray 2 shown in this file photo takes at the Ames Research Center at Moffett Field, Calif. But makers of superfast, smaller machines are finding converts in U.S. Industry.; BLACK & WHITE PHOTO

Copyright 1989