Sun Bets On High-Performance Computing For A Volume Market 2

Losing money and market share as more businesses and researchers turn to low-cost computing technologies from Intel and Microsoft, Sun Microsystems has begun an aggressive initiative to transfer new supercomputing techniques to its scientific and business products.
Sun Microsystems executives working on a military-funding supercomputing research project say they're developing a new, Java-like programming language and run-time environment that could simplify programming high-performance computing applications. They're also trying to migrate new technologies for performing on-chip calculations, moving data between chips without wires, and cooling systems more efficiently to commercial products in the next few years.

If successful, the work, which involves 90 researchers and engineers, could help differentiate Sun's products from inexpensive clusters of PCs running the Linux or Windows operating system that are used to perform more high-end tasks, says John Gustafson, a principal investigator at Sun. "The system must be cheaper or more capable than commodity Linux clusters, or why bother?" he said during a presentation at Sun Labs in Mountain View, Calif., on Wednesday.

For its most recent quarter ended Sept. 28, Sun reported revenue of $2.5 billion, an 8% decline, and a net loss of $286 million, compared with a loss of $111 million a year ago.

Five months ago, Sun, IBM, and Cray received more than $146 million in funding from the Defense Advanced Research Projects Agency to develop a new class of high-performance computers that could become the world's fastest, and address a perceived shortfall in U.S. high-performance computing research and development. Sun received $49.7 million under the High Productivity Computing Systems program. Another round of funding will be awarded in 2006. The systems being researched would emphasize ease of programming and ultra-fast transfer of data, in addition to CPU clock speeds, to achieve high levels of performance.

Jack Dongarra, a computer science professor at the University of Tennessee, says many systems that perform high numbers of math operations per second don't perform as well on moving data through the computer. Cray chief scientist Burton Smith argues that customers should measure the value of a high-performance system based on the cost of moving data, in addition to the cost of chip speed.

According to Sun chief technology officer Greg Papadopoulos, technology being developed for the Darpa program also could be useful to corporate R&D departments, which have automated fewer of their processes than other parts of American business. "It's not about building one big computer," he says. "Sun is placing a big bet that high-performance computing is a volume market, otherwise we wouldn't be doing this."