Connecting state and local government leaders
For the second time this year, an Energy Department laboratory is teaming up with industry to develop new high-performance computing techniques for studying genes and proteins.
For the second time this year, an Energy Department laboratory is teaming up with industry to develop new high-performance computing techniques for studying genes and proteins.
Oak Ridge National Laboratory and IBM Corp. have signed a one-year cooperative research and development agreement to work on scalable operating systems and biology applications.
The technology could lead to massive computers capable of hundreds of trillions or even a quadrillion operations per second, said Thomas Zacharia, director of the computer science and mathematics division at the Energy lab in Oak Ridge, Tenn. The agreement can be extended if the parties agree.
No static here
IBM is designing a supercomputer strategy called cellular architecture, in which processor cells each have their own memory and communications circuits.
The company's current RS/6000 massively parallel technology will scale to roughly 100 trillion floating-point operations per second.
When computers reach a quadrillion floating-point operations per second, or 1 PFLOPS, complications will arise 'from a civil engineering point of view,' said David McQueeney, vice president of emerging business for IBM's research division.
Systems containing tens of thousands or a million processors would be physically too large and consume too much power to be practical.
Cellular architecture is a refinement of a 1990s effort to develop so-called processor-in-memory chips, McQueeney said.
Instead of having processor chips separate from RAM, each cell would have four to 25 processors with associated memory and interconnection switches.
Such cells would dissipate much less power per million operations per second than today's supercomputer architectures, McQueeney said.
But the cellular architecture will exact a price in other ways. Although a cellular supercomputer wouldn't be a single-purpose system capable of solving only one scientific problem, its architecture would be less adaptable to a wide range of applications than today's systems, McQueeney said.
Five Oak Ridge scientists will collaborate with the IBM team on Blue Gene, IBM's nearly 2-year-old initiative to build a massive supercomputer for protein-folding studies. Such investigations could lead to new pharmaceuticals and other disease-fighting techniques.
Lab coat and blue jeans
McQueeney said IBM still will build its Blue Gene supercomputer.
Scheduled for completion in 2004, Blue Gene will have a theoretical peak speed of 1 PFLOPS.
Energy and the National Institutes of Health have been working on mapping the human genome since 1990. Scientists from federal agencies and a private company, Celera Genomics Corp. of Rockville, Md., completed a rough-draft map last year and expect to finish the high-quality sequence by 2003.
Early this year, Energy's Sandia National Laboratories signed a four-year cooperative agreement with Celera to develop life-science supercomputing technologies based on Compaq Computer Corp. platforms [GCN, April 30, Page 30].
The two labs' efforts should complement each other, Zacharia said.