A Supercomputing Star--And Why It Fell
The rise of massively parallel computers, which broke new ground in price-performance ratios, also spelled the waning influence of an American business icon.
Until the early 1990s, multimillion-dollar supercomputers built by Cray Research and Control Data were used to simulate nuclear weapons explosions at national labs funded by the Department of Defense; design complex engineered parts in aircraft and automobiles; let oil companies visualize the earth's crust; and help predict weather. Those problems require computers that excel at crunching long strings of numbers called vectors. For a time, no company's products tackled vector computing better than Cray's.
Cray Research, founded by Control Data computer scientist Seymour Cray in 1972, owned the market for vector processing during the 1970s and 1980s. In 1976, Cray shipped its first system to Los Alamos National Laboratory: The Cray-1 contained a 75-MHz, 64-bit processor and boasted a calculation speed of 160 megaflops, making it the world's fastest vector system.
"The story of Cray is, in some ways, the story of the funding sources that make it possible to build high-performance computers," says Gary Smaby, principal of high-tech venture-capital firm Quatris Fund and who covered Cray as a financial analyst during the 1980s.
Cray's stock more than quadrupled between 1976 and 1987. At its height in the mid-1980s, Cray controlled more than 60% of the market for systems priced higher than $1 million, and collected more than 20% of its sales as profits.
Sales peaked in 1994 at $922 million. But by then, Cray had become less of a factor in setting technology directions. Steve Oberlin, Cray's former chief architect for massively parallel systems, and now CEO of Linux scalability firm Unlimited Scale Inc., says the company become "enamored with its own technology."
With the end of the Cold War, the government scaled back funding for supercomputer projects. The market was changing in other ways, too. During the late 1980s, the military's Defense Advanced Research Projects Agency (Darpa) R&D arm declared that parallel systems built with off-the-shelf components could do the work of Cray's CISC-based vector processors. The debut of IBM's RS/6000 SP in the early 1990s helped bring parallel processing to the fore of supercomputing. (In 1989, Seymour Cray founded a new company, Cray Computer Corp., which closed in 1995).
Cray shipped two massively parallel computers, but the company's adherence to proprietary components put it at a cost disadvantage. The Cray T3E, which shipped in 1996, used a commercially available Compaq processor, but its operating system, cabinets, cables, memory, and input/output system were custom-made. The machine scaled to thousands of CPUs, but it was expensive compared with commercial off-the-shelf systems. "It was really the huge step away from shared memory to distributed memory that Cray and lots of people resisted," Oberlin says. "Clusters today represent another step along the price-performance curve."
Meanwhile, Darpa kept cooling its supercomputing funding. In 1996--the same year Seymour Cray died after a car crash--Silicon Graphics Inc. bought Cray Research for about $740 million. Last year, Tera Computer Co. in Seattle picked Cray up from SGI for less than $100 million, and changed the company's name to Cray Inc.
Other vendors' supercomputing efforts had tanked as well. Intel exited the business in 1996, unable to sustain the high cost of manufacturing specialized processors. Thinking Machines Corp. fell into bankruptcy before Oracle bought its technology in 1999. Even Compaq, enjoying success in the supercomputing field supplying life-science company Celera Genomics Group, says it can't afford to keep building its Alpha processor.
Says Debra Goldfarb, a supercomputer analyst at International Data Corp., "We've always been looking for that killer application for parallel computers. Genomics has won."
A Cray spokesman says the company's bio-informatics efforts center around building system software with advanced management tools, while Compaq controls the market for computational computers, and Sun Microsystems owns the data-analysis market. Reports say the National Security Agency, the government's cryptologic organization, is providing financial support for the development of Cray's next-generation SV2 supercomputer.
About the Author
You May Also Like
2024 InformationWeek US IT Salary Report
May 29, 20242022 State of ITOps and SecOps
Jun 21, 2022