LAST week IBM confirmed the existence of the rumored Baby Blue Gene or Blue Light supercomputer a scaled-down version of the Blue Gene cellular architecture officially named Blue Gene/L.
In partnership with the Department of Energy''s National Nuclear Security Agency''s Lawrence Livermore National Laboratory, IBM said it would jointly design a new supercomputer based on the Blue Gene technology. The new machine will be 15 times faster, 15 times more power efficient, and take up 50 times less space per computation than current supercomputers, IBM said.
Blue Gene/L is expected to operate at about 200 teraflops, which IBM said is larger than the total computing power of the top 500 supercomputers in the world today. It is expected to be completed by 2005.
IBM said it would continue to build the petaflop-scale Blue Gene that the company first announced in December 1999. While Blue Gene is being designed for the very specific purpose of protein folding simulation, researchers at the national laboratories plan to use Blue Gene/L to simulate a far broader range of physical phenomena, such as the aging of materials, fires, and explosions.
Additionally, IBM and LLNL expect the architecture for Blue Gene/L to be more easily adaptable to commercial applications, eventually bringing the machine out of the lab and into business and industry. IBM is actively pursuing a partner to contribute to the development of a commercial version of the machine.
IBM initially had no plans to move toward a commercial version of the Blue Gene architecture, and described the project as a grand challenge that would build upon its previous research projects such as the Deep Blue chess-playing computer. However, As the project evolved, we realized there was an opportunity to make the architecture applicable to a broader range of applications, said Bob Germain, a manager in the computational biology center at IBM Research who heads up the science and application portion of the Blue Gene project.
BlueGene/L represents a major step along the path to getting to our eventual goal of building the even larger computer as part of the same project, said Germain. While Blue Gene was originally expected to take four to five years, Germain declined to pin down a completion date for the petaflop machine.
We still intend to get to that petaflop, but it may be that we stretch out the schedule somewhat, said Germain.
IBM will focus on the hardware side of the project and will develop the single-purpose protein folding applications for the petaflop computer, while LLNL will develop the broader applications it intends to run on Blue Gene/L and the conceptual design of the equipment, according to David Schwoegler, the Accelerated Strategic Computing Initiative program representative at LLNL.
IBM has already partnered with Lawrence Livermore to deliver the 7.2-teraflop ASCI White machine at LLNL, which currently ranks as the world''s fastest supercomputer.
Schwoegler said that BlueGene/L would take the workload off the ASCI machines that are doing the button-to-bang work of nuclear weapons simulation. The Blue Gene architecture is expected to perform better on computational problems that can be divided to run on tens of thousands of processors.
The joint work on Blue Gene/L is currently being performed under the existing IBM/ASCI contract, Schwoegler said. Any revenue-sharing plans for the eventual commercialization of the technology would require new contractual arrangements, and those are in the works.