Around the turn of the century, Eric Kronstadt, IBM's director of big computing systems and research, and his Yorktown...
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
Heights, N.Y.-based team decided to set a grand challenge for themselves.
They evaluated the two-pronged process science traditionally deployed -- theory and experimentation -- and figured that supercomputers could help give birth to a third approach: simulation.
Supercomputing allows researchers to test more theories, try more designs and simulate more conditions. And one of the most significant realms within the scientific community that Big Blue thought could take advantage of supercomputing was computational biology.
The five-and-a-half-year pursuit of that challenge finally bore fruit Monday with the launch of the Watson Blue Gene system, nicknamed BGW, which IBM said is the world's most powerful privately owned supercomputer.
BGW's first task will be to produce a full-scale simulation of the folding of a complex protein, a task that, if completed, will represent a significant step forward for biology.
"The way [a protein] folds is very important in trying to understand how it functions," Kronstadt said. "We want to simulate the first microsecond of the lifespan of folding to see how it takes place, and we need a supercomputer [to do that]."
With a processing speed of 91.29 teraflops, the system is expected to clock in as one of the top three supercomputers in the world.
BGW is comprised of 20 refrigerator-sized racks, less than half the size of conventional systems of comparable power and has three times the performance. Each rack contains 2,000 dual-core Power chips; each with 4 MB of embedded RAM, and can produce as much as 5.6 teraflops of computing power at peak performance.
When Kronstadt's team began to develop the road map for the BGW, it estimated it would take as much as a petaflop (1,000 teraflops) of power to run a protein folding simulations. But with help from the staff at the Lawrence Livermore National Laboratory working on its sister machine, the Blue Gene/L, the team was able to funnel its resources in a much more efficient manner.
"We were a little bit clever about what we needed to do to understand the process of protein folding," Kronstadt said. "We refined our capabilities and estimates of what we need so we're pretty comfortable with this machine."
The BGW is a big boy, but Kronstadt insists cooling won't be an issue. It has its own air-cooled data center, and IBM isn't worried about potential overheating because the BGW is anywhere from three to 10 times less power hungry than comparable super boxes, and thus much less likely to tax the room with excessive electrical concerns that can lead to heating nightmares.
Of the 2,000 processors in each rack, 128 will run Linux and serve as the interface for programmers, while the rest will run what Kronstadt describes as a stripped-down version of Linux. Linux-based MPI, or message passing interface, is the operating kernel that will run on top. IBM hopes to use the system to explore how it might enable progress in a range of fields, from technical -- including life sciences, hydrodynamics, materials sciences, quantum chemistry, molecular dynamics and fluid dynamics -- to business applications.
The BGW is now the second-most powerful computer IBM is working on, behind the Blue Gene/L, which currently supports thirty-two 2,000 chip racks and is expected to beef up to 64 racks by the end of 2005.
Let us know what you think about the story; e-mail: Luke Meredith, News Writer