The company plans to announce the concept Monday with its multi-year, three-phase "adaptive supercomputing" plan, said Jan Silverman, senior vice president of corporate strategy.
"No single processor architecture can best execute all programs," Silverman said. "You really need a combination of processor architectures to execute as efficiently as possible. And you have to hide that complexity from the users."
Cray faces competition from competitors such as IBM, Hewlett-Packard, Sun Microsystems and Dell, whose systems work for mainstream computing tasks and therefore sell much more widely. And the company could use a.
On Friday, it reported preliminary results of a $65 million net loss on revenue of $201 million, and the revenue and net loss of 2004 could worsen by as much as $3.3 million because revenue from a product development contract may have been recorded improperly, the company said. The moves sent the company's stock down 11 percent to $1.82. But the company also said it expects 2006 revenue to grow about 5 to 15 percent above 2005 levels.
The first phase of Cray's product overhaul begins with separate systems tuned for different types of supercomputing jobs but sharing some hardware elements, Silverman said. Those products should emerge in 2007.
The second phase, roughly two years later, combines the systems into a single chassis, with separate blades for specific types of computing problems. In the third phase, control software will automatically route computing tasks to the best-suited hardware available in a system.
One of the four chip architectures is Cray's traditional stronghold, "vector" processors that can execute particular types of mathematical problems very quickly. A second type are the more ordinary, typically placed in a large number of networked computers. The third are field-programmable gate arrays (FPGAs), chips that can be reconfigured on the fly to run specific programs very quickly. And the fourth are multithreaded chips from and assumed the better-known company's name.