IBM on Friday looked to take its Blue Gene supercomputing research project to the next level, announcing a partnership with the Lawrence Livermore National Labs to create a machine 15 times faster than its current fastest supercomputer.
Described as a major expansion of the Blue Gene project, the Blue Gene/L machine is expected to run at 200 teraflops, or 200 trillion operations a second. The company hopes the new architecture will eventually have more appeal to a broader set of applications outside of the scientific world.
"We anticipate that because this architecture represents a new price-performance point, as well as a breakthrough in power consumption, it will be more adaptable to a broader range of commercial applications eventually," said Bob Germaine, who manages the science and application effort in the Blue Gene Project at IBM's Computational Biology Center in Yorktown Heights, New York.
The new joint development effort will be part of the Department of Energy's National Nuclear Security Agency's Accelerated Strategic Computing Initiative, the project under which IBM and Lawrence Livermore delivered the ASCI White machine, now in use at Lawrence Livermore.
At least initially, researchers at the labs expect to use Blue Gene/L, which will not be completed until 2005, to simulate physical phenomena such as the ageing of materials or fires and explosions.
"I see the (ASCI White and Blue Gene/L) projects being complimentary. Rather than building Blue Gene on our existing SP2 products, like we did with ASCI White, we are taking a new technology approach in order to find better price performance and power performance points," Germaine said.
Some of these new technical approaches, centered around exploring cellular systems and lower-power-consuming processors, should result in systems that are significantly less expensive to run, maintain and house, thereby making them more appealing to commercial businesses.
The Blue Gene/L systems will consume 15 times less power and space than today's faster supercomputers. For instance, the space needed for ASCI White is about the size of two basketball courts, while Blue Gene/L will only need about half a tennis court, according to a company spokesman.
Blue Gene researchers realised the system's architecture could serve as a foundation for delivering architectures more suitable for commercial applications, while still delivering a machine to on its original plan of protein science simulations.
"Patnering with Lawrence Livermore is a key part of our strategy, as they bring important application and design expertise to the project," said Mark Dean, vice president of systems at IBM Research.
What will allow Blue Gene/L to run applications faster is 'data-chip cells' specifically optimised for data access. Each chip will include two processors, one for computing and one for communicating. Each data chip cell will work on a small part of a bigger problem, company officials explained.
IBM sees Blue Gene/L as a necessary step towards achieving its even longer-term goal of building a pedaflop (1 million gigaflop) machine, which would have the power of a million higher-end desktop computers.
For more information about the Blue Gene Project and other projects at IBM Research, users can go to www.research.ibm.com.