Cray Research Inc. is used to being on top. Since 1976, when founder Seymour Cray first shipped his ultrafast Cray 1, the company has been nearly synonymous with supercomputing power. Indeed, Cray Research made an astonishing 65% of the advanced supercomputers in use today, and engineers and scientists still regard its technical prowess with awe.
That's what makes a new advertising campaign by brash upstart nCube Corp. so startling. The ads prominently feature an assortment of outmoded technology, including a stone tablet, a vintage manual typewriter -- and a Cray 1.
BROKE SLOWLY. Cray's usually serene chief executive, John A. Rollwagen, is annoyed by the persistent criticism that powerhouse Cray Research is late getting into the market for massively parallel processors. "I guess we'll give up," he growls sarcastically. But he doesn't deny that Cray is behind in MPPs. "Do I wish we had a product on the market today?" he asks. "Yes. Do I wish we had started a little earlier? Yes. But being first doesn't take the day. We're going to do it best."
To get there, Rollwagen last year launched an MPP research team run by Steve Nelson, who just finished designing the company's newest supercomputer, the Y-MP C90. It can do some parallel processing, which splits calculations over many processors, but it will be used mostly to handle data sequentially, one calculation after another. By 1993, Cray's MPP research will bear its first real fruit: an MPP add-on that will gang together as many as 1,024 microprocessors.
Cray's long-term plans are even more ambitious: It wants to put a new twist on MPP design by trying to make all those microprocessors swap information faster than other MPP computers. To help accomplish that, Cray Research plans to shorten the machine's internal wiring and jam the processors closer together. The method is more expensive, but Rollwagen says that as a result, Cray Research by 1997 will be the first company to make a computer that can sustain 1 trillion operations per second, the vaunted teraflops.
To win the teraflops race, it will have to beat such competitors as Intel, nCube, and Thinking Machines. And Cray's late start doesn't help. The company resisted MPP until recently because the technology was a big departure from Cray's roots. New technology "is always a challenge for any company with a mature product," says F. Ron Bailey, director of aerophysics at NASA Ames Research Center.
It's still hard to think of a Cray supercomputer as a mature product. Throughout the last decade, Cray machines radically changed the course of science and engineering, allowing scientists to simulate nuclear blasts in outer space and engineers to fake car crashes.
NOT SO HOT. The key to the technology was Seymour Cray's work in vector processing, a way of zapping information through computer chips 10 times faster than was previously possible. Today, that's not fast enough: "The days are numbered" for vector machines, says David M. Tolle, research manager for emerging technologies at Shell Oil Co.
Now, the question is how healthy Cray Research can remain until it moves into MPP. Rollwagen says that because it's hard to develop MPP software, vector computing will remain a strong market for many years. That will help Cray Research report steady increases in revenue and earnings, says Barry F. Willman, an analyst at Sanford C. Bernstein & Co. In 1992, he projects, Cray will earn $ 133 million on sales of $937 million, vs. 1991 projected earnings of $111 million on sales of $850 million.
Meanwhile, Seymour Cray, now 66, left Cray Research in 1989 to set up Cray Computer Corp. He's working on a vector machine that uses ultrafast but expensive gallium arsenide chips. And he has no plans to develop MPP. "Plenty of others are doing it. We want to make a unique contribution," explains Cray Computer President Neil Davenport. The grand old man of supercomputers, for one, doesn't think MPP is the only way to go.