IBM Teams With DOE To Build Supercomputer 30X Faster Than BlueGeneIBM Teams With DOE To Build Supercomputer 30X Faster Than BlueGene

IBM is teaming up with the U.S. Department of Energy to build a supercomputer they expect to be 30 times more powerful than the fastest machine running today.

The Office of Science and the National Nuclear Security Administration (NNSA), both of which are under the umbrella of the Energy Department, will share the cost with IBM for the five-year, $58 million research and development effort, according to IBM. The project will build on the capabilities of the IBM BlueGene/L system, the most powerful supercomputer in the world today. Installed at the Lawrence Livermore National Laboratory in Livermore, Calif., BlueGene has a Linpack performance of 280.6 teraflops.

A second BlueGene system, the eServer BlueGene Solution system, installed at IBM's Thomas Watson Research Center and running 91.20 teraflops, is the third most powerful supercomputer in the world.

The new project would eclipse both of these machines.

"This is very significant," says Jack Dongarra, a professor of computer science at the University of Tennessee and one of the authors of the Top500 supercomputer list. "Something much faster that will allow scientists to do things they can't do today, even on the fastest computers at Livermore. This will be used to solve some of the most challenging problems we face."

Researchers hope to use the new supercomputer to monitor the aging process of the U.S. nuclear stockpile, speed genome sequencing, and model climate changes, according to IBM. The company will work on the development with scientists from Argonne National Laboratory and Lawrence Livermore.

The NNSA and the Office of Science will each contribute $17.5 million, and IBM will contribute $23 million.

"Supercomputers are crucial to the continued success of the NNSA's science-based efforts to keep the U.S. nuclear weapons stockpile safe, secure, and reliable without underground testing," said NNSA Administrator Linton Brooks in a statement. "Computing at these scales will enable predictive simulations that allow researchers to understand how complex physical, chemical, and biological systems behave over time. Previously, it was only possible to get brief snapshots on a smaller scale."

Charles King, principal analyst at Pund-IT Research, says the project doesn't have any short-term implications for businesses, but in the long run there will be lots of benefits for large companies.

"Eventually, the price of supercomputers will drop and the work that IBM is doing now will have broad implications down the line," he says. "As this level of computing becomes more common, people will think of ways to use it."

While a super high-speed computer is interesting, most businesses will benefit more from supercomputing clusters, says Tony Asaro, a senior analyst with the Enterprise Strategy Group. "The question is what's the price going to be for that performance?" asks Asaro. "We have to take a look at the fact that there are initiatives that are happening with clusters. We're getting scalable, low-cost computer power [with clusters]. With clusters, I can use whatever components I want. I can use whatever operating system I want. I'm not bound to a specific vendor. It gives me more options and more freedom."

Computing performance will make major leaps forward in the next five years, says Dongarra, who predicts machines will pass the petaflop barrier (1,000 trillion operations per second) in that time frame. Los Alamos is working on the Roadrunner hybrid supercomputer that combines cell chips with a base cluster that runs on the Linux operating system and uses IBM System x3755 servers based on AMD Opteron technology. That system is expected to be up and running sometime in 2008.

[UPDATE: Nov. 16 In the last paragraph, a reference to Roadrunner was clarified to indicate that the hybrid supercomputer is based on AMD Opteron technology.]