Supercomputer 'starter kit' ready

The "starter kit" for the National Science Foundation's most powerful supercomputing resource became operational weeks ahead of schedule.

When the Terascale Computing System at the Pittsburgh Supercomputing Center becomes fully operational in October, it will be able to perform 6 trillion calculations per second at its peak. The initial capability now available is about half a teraflop, or 500 billion calculations per second.

"We were committed to having significant initial operating capability by Feb. 1, and we beat that by a few weeks," said Robert Borchers, director for advanced computational infrastructure and research at NSF.

Compaq Computer Corp. will start shipping the final nodes to bring the system to its full capability in May and complete the deliveries in October.

NSF awarded the system to the Pittsburgh Supercomputing Center in August 2000. The supercomputer, which is the fastest available for civilian research, is funded at $36 million, plus $9 million over three years for operating costs.

NSF's strategy is to initially allocate the machine for small projects, Borchers said. The allocation committee will begin awarding use of the system in April, he said. Some of its potential uses are for research on better storm, climate and earthquake predictions; more efficient combustion engines; and better understanding of chemical and molecular factors in biology.

"What we're giving people now is enough time to move code, check the answers to test problems and make sure the software they need is available," Borchers said. "We'll ramp up people's usage as the processing power increases this summer."

The "starter kit" has 64 interconnected Compaq ES40 AlphaServers, Borchers said. By October, those will be replaced by more than 682 faster AlphaServers.

On Jan. 18, NSF also solicited proposals for a second, distributed terascale facility that will broaden civilian research capabilities. Proposals are due April 19, and Borchers said he hopes to be able to submit the $45 million award for approval by National Science Board at its May meeting.

"We are asking proposers to propose at least two sites to get some sense of how you compute across distances," Borchers said.