IBM, Energy raise supercomputer scale

To solve computational problems in critical areas where current computing technologies fall short, the Energy Department's Oak Ridge National Laboratory and IBM Corp. will collaborate on building machines dramatically faster than today's most powerful supercomputer.

That kind of computing power can be used to lead to breakthroughs in studying climate, advancing the field of nanotechnology and gaining a better understanding of gene sequences and diseases, officials from Oak Ridge and IBM said in announcing the cooperative research and development agreement Aug. 22.

IBM's Blue Gene research project, which is based on its cellular architecture supercomputer design, will serve as the foundation for collaboration. Unlike today's computers, cellular servers will run on chips containing "cells"— processors that contain memory and communications circuits. Each processor forms a cell with built-in memory, communication and input/output ability.

Cellular architecture will help scale computer performance from teraflops(1 trillion calculations per second) to petaflops (1,000 trillion calculations per second). The fastest existing computer — ASCI White, unveiled by IBM last week — can perform about 12 trillion calculations per second.

Petaflops of computing power could be used for everything from climate and weather modeling to protein folding research that aids in understanding diseases, said Joe Jasinski, manager, computational biology center for IBM Research.

"The whole idea behind the Blue Gene project is to build machines...that are orders of magnitude" faster than ASCI White to advance those research areas that are shortchanged by today's computational limitations, Jasinski said. "There's too much computation to be done in a reasonable amount of time...and the common requirement is that they are limited in scope by the amount of computing power we can bring to bear on them."

Scientists from Oak Ridge will work with IBM on programming the new supercomputer to analyze proteins.

With current technology, machines capable of the computing power needed to advance nanoscale science and climate dynamics would quickly become unbearably large, costly and inefficient. "The technology for the Blue Gene hardware is quite extraordinary and represents a fundamental shift in the way computers are built, “ said Thomas Zacharia, director of Oak Ridge's computer science and mathematics division.

However, greater power means a more pressing need for applications that use it, which Zacharia said is like having a sports car, but not knowing how to drive it. "As we're developing and deploying these terascale machines, we need to learn, develop and deploy application software to take advantage of it."

He said protein-folding research is basically at a standstill with current computing power and Blue Gene promises to deliver the advances that can "tackle the fundamental questions and help develop drugs and help humanity in general."

It shouldn't be long before scientists will begin reaping the benefits of the yet-unseen computing power. "The goal is to have the first of these machines functioning by 2004," Jasinski said, adding that chip design is under way and hardware testing will begin next year.