IBM helps customers get a supercomputer on the cheap

IBM ratcheted the world of supercomputing up a few notches in June with the Blue Gene/P, a system nearly three times as fast as its predecessor at a cost of $1.3 million per rack.

But in anticipation of the Blue Gene/P, IBM dropped the price of the Blue Gene/L, to about $800,000 late last year and prompting sales of Blue Gene/L to more than doubling the first half of this year, compared to the second half of 2006, says Herb Schultz, IBM’s deep-computing marketing manager. At its highest price, the Blue Gene/L cost $1.3 million per rack, same as the P’s current price.

“It’s still a very viable platform,” Schultz says. Among universities, “we’ve had some really big sales, RPI and Stony Brook, for instance,” he says.

Another buyer was the University of Alabama at Birmingham, which begun using a Blue Gene/L a month or two ago to design drugs that could treat clogged arteries, neurological diseases or certain types of cancer.

UAB is not a minor player in research, doing more than $225 million worth of work for the National Institutes of Health each year. But it was reluctant to splurge on a supercomputer until the recent price drop.

“We knew the L was a model near the end of its production, and we were able to secure a much better price on that than we would on the newer model,” says Richard Marchase, vice president for research and economic development at UAB. “For our purposes, the L had plenty of capacity.”

UAB tripled its computing power in computational biology and molecular simulations with the purchase. The supercomputer will shorten the yearslong process of developing drugs targeted at specific protein structures, Marchase explains.

In computational biology, UAB researchers will use the supercomputer to examine data about proteins and find protein structures that are thermodynamically stable, he says. Once those structures are identified, which could happen in six to eight months, researchers can begin figuring out what kinds of small molecules could interact with protein structures in ways that cure diseases, he says.

“The increase in speed that we were able to purchase with the Blue Gene is allowing us to go through these iterations,” Marchase says. “These processes are very iterative,” he says, requiring researchers to study individual structures and improve upon them incrementally over many steps.

The Blue Gene/P can perform 13.9 trillion operations per second, compared with 5.6 trillion for the Blue Gene/L purchased by UAB.

IBM doesn’t want the Blue Gene/L’s late-in-life sales increase to last forever. Schultz says IBM is aiming to transfer existing customers to the Blue Gene/P, which delivers more power per dollar and per watt.

The Blue Gene/P has four publicly announced customers, including the U.S. Department of Energy and the Max Planck Society for the Advancement of Science.

IBM expects to announce additional signings throughout the summer and to eventually find a customer to buy a petaflop system composed of 72 Blue Gene/P racks, according to Schultz. A petaflop machine could perform 1 quadrillion mathematical calculations per second.