Cray fills need for computer speed
Special to The Seattle Times
The world knows Seattle is home to leaders in software, coffee and, until recently, airplanes. But time machines?
Headquartered in offices at Pioneer Square, Cray makes computers its executives like to call time machines because their product emphasizes speed, an unimaginably huge number of calculations performed nearly instantaneously. The machines, part of a powerful class called supercomputers, also save the "lives" of crash-test dummies, have been involved in intelligence work for the government and are working furiously to recapture the lead position in scientific supercomputing research for the United States, an accolade some say was recently conferred on the Japanese.
Cray, a name well-known in the history of computing, has had a tortuous odyssey that landed it in Seattle in 2000. That was when Seattle-based Tera Computing, a developer of supercomputers, bought select assets from financially troubled Cray Research, which had merged in 1996 with Silicon Graphics Inc. (SGI). Tera renamed the surviving company Cray Inc. and set out to reverse the decline resulting from SGI's move to halt all new Cray product development in 1996.
It had been a long downswing since Seymour Cray, sometimes called the "father of supercomputing," founded the company in 1972, locating manufacturing and research in his hometown of Chippewa Falls, Wis., and headquarters in Minneapolis.
From those Midwestern towns, Cray competed with the likes of IBM and other computing heavyweights. (The Cray company still has offices in Chippewa Falls and Minnesota.)
Even before starting the company, Seymour Cray had designed early supercomputers, machines shaped like a room, surrounded by glass walls with wood trim and big enough to stroll inside and sit down. Created in the 1960s, they were about as fast as Intel's Pentium 4 processors, the chips that run some of today's desktop computers.
A new product
One definition of supercomputers considers them to be the fastest computers at any given point in time. The X1, Cray's newest product, expected to be commercially available by year's end, is a supercomputer 25,000 times faster than a Pentium 4.
If that figure isn't tough enough to imagine, Cray's goals really boggle the mind. It plans to produce a supercomputer with sustained petaflop performance by the 2010 timeframe. That means the computer can do as many calculations as a one followed by 15 zeros — in one second.
While Cray executives sound confident that its supercomputers can soon surpass the world's fastest in Japan, the company's position in the commercial sector is threatened by increasingly powerful competitive forces, most notably the personal computer.
Cray's rivals in supercomputing include makers of "cluster supercomputers," including mainstream PC makers Hewlett-
Packard and Dell Computer.
While there are some problems that only Cray's type of product can solve, cluster makers are nibbling away at a portion of Cray's market.
The cluster makers' approach developed in the early 1990s as companies, aided by increasingly cheaper computer components, learned they could cobble together off-the-shelf computers into clusters. That allowed them to build their own supercomputers for a fraction of the price of a custom-made version.
"You could say that we followed the example we saw with customers," said Steve Joachims, marketing director for Hewlett-Packard's high performance computing division.
HP and other computer makers began to actively market cluster supercomputers, eating away at the commercial market for Cray computers, including customers in the automobile and aerospace industries.
Last year HP, together with Compaq, accounted for about 40 percent of the top 500 most powerful supercomputers, according the Top 500, a list put together twice a year by researchers in Germany and the United States. A new Top 500 list, released last week, featured cluster computers in the top 10 for the first time ever.
The price of an HP cluster computer varies, but can start at less than $100,000 and go up to a couple million — the starting point for Cray's newest supercomputer.
But there are still certain problems that only vector supercomputers, the kind Cray makes, can solve. One difference in Cray's products — and the reason executives there call them time machines — is that they focus on making calculations faster than other types of supercomputers, an important consideration for certain types of problems.
For example, wildfire observers hope to use supercomputers to predict the behavior of fires so they can control or stop them. Currently, no computer can analyze all the data that determine where a fire will go before the fire actually gets there.
"By the time they simulate it, the wildfire has gone past the simulation," says Burton Smith, chief scientist at Cray. "The objective often is to simulate faster than you can do it in reality."
Cray's computers are also ideal for various applications required by the government, which has historically used supercomputers for nuclear-weapon design and intelligence work, including cryptoanalysis, or code-breaking.
"The passage of the Homeland Security bill and the defense bill are favorable to the kind of business we're in," says Jim Rottsolk, chairman, president and CEO of Cray. Rottsolk founded Tera Computing with Smith.
The government, in fact, is Cray's largest customer, and even the company's competitors recognize how valuable that customer is. "I don't see them going away because of their strategic importance with the government," said HP's Joachims.
In October, the Department of Energy's Sandia National Laboratories, in Albuquerque, N.M., awarded a $90 million contract to Cray for the development of a new supercomputer expected to be seven times more powerful than the one the lab uses.
The Department of Energy also plans to use a Cray supercomputer, among others, to try to catch up to the Japanese, who recently began employing a supercomputer made by NEC that's now the fastest on earth.
Called the "earth simulator" because it's used for climate studies, the Japanese supercomputer shook up the U.S. scientific community because it marked the first time that the world's fastest computer was outside this country. Cray's X1, in its largest form, could exceed the performance of the earth simulator, Rottsolk says.
But Cray hasn't written off the commercial sector and envisions innovative uses for its supercomputers. "People are just beginning to use [supercomputers] to design drugs," says Rottsolk.
Instead of using a "wet lab," or an actual laboratory, drug researchers can use supercomputers to simulate experiments, adding many more variables than otherwise might be possible. A supercomputer could model diseases and work out effective medicines or vaccines, Rottsolk says.
While HP's Joachims says his company and others now serve the bulk of automobile and aerospace customers, some of those companies still turn to Cray. Boeing recently bought a Cray vector supercomputer, and Ford Motor recently purchased clustered Dell computers, which Cray pulls together for customers.
While supercomputers are expensive — the X1 from Cray will average $8 million but can cost as little as $2.5 million to more than $100 million — some commercial users can make them pay off.
Take crash analysis in the automobile industry.
"How many cars do you run into a brick wall at what speed, and if the dummy is 2 inches taller, what happens? So if you can simulate that, you get a much finer grain," says Lori Kaiser, vice president of marketing and strategic planning at Cray.
For all the visions of how its product can be used, Cray has felt the sting of the economic slowdown, as its customers defer spending on expensive equipment. Still, for the quarter ended Sept. 30, the company managed to record revenue of $42.1 million, up 43.1 percent from the same period in 2001, while posting a profit of $2.1 million, compared with a net loss of $10.8 million the previous year.
The stronger numbers have helped Cray's stock grow from less than $4 a share to more than $6 since mid-September. "There's a realization from a financial viability point that we're past our real challenges," Rottsolk says.
Last year, Cray plowed money into research and development, spending more than 40 percent of its revenue on R&D, in the hopes of making up for the lost time when Cray's previous owner, SGI, cancelled product development.
While SGI may have contributed to Cray's declining market presence, it also gave Rottsolk and Tera an opportunity he hadn't dared to dream about. "We never envisioned we'd have the cash to buy them," said Rottsolk. "They were the company we wanted to emulate."
Since acquiring the assets, the resulting company has focused on developing the X1. Even without a new product, the company was profitable in the past three quarters, says Rottsolk.
Those factors, combined with having a new product due to be released on schedule, have been winning over the investor community, says Alan Davis, an equity analyst with McAdams Wright Ragen.
"There's a lot of pent-up demand, especially initially from government customers," he says.
Nancy Gohring is a freelance writer in Seattle who specializes in technology coverage.