The new machine, to be used at Japan's National Institute of Advanced Industrial Science and Technology (AIST) should be capable of sustained performance of 17.2 trillion calculations per second, IBM plans to announce on Tuesday. AIST's Computational Biology Research Center will use physics simulations to investigate how genetic information encoded in DNA becomes a string of building blocks folded into the complicated three-dimensional shape of a protein.
This protein-folding problem was the original challenge that led IBM to launch the Blue Gene programme five years ago. Protein folding is important for understanding biochemistry, and important to using computers to design drugs.
The AIST system, a relatively modest machine consisting of four racks of equipment, is the fourth Big Blue has sold so far. IBM also has sold a massive 64-rack machine to Lawrence Livermore National Laboratory, a six-rack machine for the Astron radio telescope project, and a one-rack system for Argonne National Laboratory.
The Blue Gene programme has been making its mark on the supercomputing world. Two Blue Gene/L systems placed in the top 10 on most recent list of the world's 500 fastest supercomputers. The AIST system is twice the size of the current No. 8 model.
In addition to the four systems IBM has sold, the company also is building a 20-rack Blue Gene/L system at its Thomas J. Watson Research Center in Yorktown Heights, N.Y., said William Pulleyblank, director of exploratory server systems for IBM Research. In addition, the Mayo Clinic is renting processing power from another Blue Gene/L IBM in Rochester, Minn.
Blue Gene/L has begun its move beyond the research phase, and an executive within IBM's Systems Group -- whom Pulleyblank refused to identify -- now is in charge of making Blue Gene/L a product that customers can buy.
"We are doing the commercialization on the Blue Gene that's been launched," Pulleyblank said, and IBM has a manufacturing facility building Blue Gene/L systems in Rochester.
IBM is well on its way to developing a successor to Blue Gene/L, Pulleyblank said. That system is likely to arrive about two years after the early 2005 formal arrival of Blue Gene/L, he said. He described the successor as "an evolutionary follow-on from the Blue Gene/L platform," though with more radical changes than simply faster processors.
IBM also is working on other designs called Blue Gene/C and Blue Gene/P, the latter designed to reach performance of 1 quadrillion calculations per second, or 1 petaflops. The Blue Gene/L being built at the Livermore lab is expected to perform at 30 percent of that speed -- 300 trillion calculations per second, or 300 teraflops.
The Blue Gene/L machines run Linux and use processors from IBM's Power family.
