by CIO Staff

IBM, ClearSpeed Team Up for Supercomputers

News
Jun 27, 20063 mins
Data Center

IBM made a move to get ahead of rival Hewlett-Packard in the high-performance computing market Tuesday, announcing a partnership with semiconductor maker ClearSpeed Technology.

ClearSpeed, of San Jose, Calif., designs microprocessors that sit alongside standard CPUs and accelerate their performance by handling the most compute-intensive tasks.

IBM will supplement its System Cluster 1350 with a ClearSpeed Advance accelerator board, featuring the CSX600, a 96-core processor that handles 25 billion floating point operations per second (gigaflops) of computation while consuming less than 10 watts of energy.

Each board has two CSX600 processors, for a total of 50 gigaflops performance at a power cost below 25 watts. A typical server chip from Intel or Advanced Micro Devices (AMD) uses between 60 and 100 watts.

The ClearSpeed “co-processors” are efficient because they have very low clock speed, running at just 250MHz compared to commercial server chips that run at 3GHz.

The extra speed gives IBM an advantage in selling supercomputers to high-end users in the fields of finance, oil and gas exploration and life sciences.

In turn, ClearSpeed will get an opportunity to sell its “co-processors” in markets beyond its traditional users in the “Top 500” supercomputers, ranked at sites like www.top500.org.

“We will deliver our technology for production level systems in commercial and industrial markets. There is much more sales volume for business there than just the 500 largest supercomputers,” said Mike Calise, head of global business development for ClearSpeed.

Those new customers will use ClearSpeed processors to build hybrid clusters, a combination of standard servers with extra memory and floating point accelerators.

The result is a system that delivers high performance for lower cost than traditional supercomputers from Cray or mid-range servers from Sun Microsystems, said Peter Ffoulkes, U.S. marketing manager for ClearSpeed.

“Standard servers are inexpensive and they offer scaling and energy efficiency. So now we just add memory and accelerators to tune it. It’s like a hybrid car that uses a gas engine at some times and an electric engine at others,” Ffoulkes said.

One of the first users is the University of Bristol in England, which will devote three clusters provided by IBM and ClearSpeed to research in climate modeling, astrophysics and pharmaceutical discoveries.

The system has 636 IBM System x servers, each equipped with two dual-core AMD Opteron processors and ClearSpeed Advance accelerator boards, drawing data from 100 terabytes of disk storage.

Universities are not the only customers demanding supercomputers that are not only fast but also power efficient.

Cray announced on June 15 it had agreed to build a US$200 million computer for the U.S. Department of Energy’s Oak Ridge National Laboratory in Oak Ridge, Tenn. Powered by Opteron processors, this computer will run at 100 teraflops by late 2006, 250 teraflops by late 2007 and 1 petaflop by 2008.

-Ben Ames, IDG News Service (Boston Bureau)

Check out our CIO News Alerts and Tech Informer pages for more updated news coverage.