In Pursuit of Power

NEC is not going to let IBM wrest the title for the fastest supercomputer without a fight.

NEC Corp. is not going to let IBM wrest the title for the fastest supercomputer without a fight.

With the SC2004 supercomputing show less than a week away, NEC has unveiled the SX-8 vector computer, a system with a peak performance of 65 teraflops, or 65 trillion calculations per second, company officials claim. That performance almost doubles that of one of the top machines in IBMs Blue Gene project, which, in late September, peaked at 36.01 teraflops, IBM officials said.

At the show in Pittsburgh, the Top500 organization will release the list of the 500 fastest supercomputers. Since 2002, Japans giant 5,120-processor Earth Simulator supercomputer, built by NEC, has sat atop the list, with a sustained peak of 35.86 teraflops.

The Earth Simulator stunned U.S. scientists and politicians, who feared the country was losing its technological edge and pushed for programs from computer makers, including IBM, Cray Inc. and Silicon Graphics Inc., to build faster supercomputers.

IBMs Blue Gene project is aimed at creating powerful supercomputers that are smaller and consume less energy than their counterparts, said officials in Armonk, N.Y. A Blue Gene machine currently under development by IBM at Lawrence Livermore National Laboratory in Livermore, Calif., will reach a peak of 360 teraflops, officials said.

/zimages/1/28571.gifClick here to read about the TOP500 Projects 2004 listing of the worlds fastest supercomputers.

NECs SX-8 is designed to carry more punch in a smaller package, too. According to officials with NEC, of Tokyo, the SX-8 is 25 percent smaller and consumes 50 percent less power than conventional supercomputers, thanks to the ability to put processors and memory on a single module. A single node carries as many as eight processors; the largest multiple-node configuration includes 512 nodes.

But NEC and IBM are not alone in the power pursuit. SGI last week said the Columbia supercomputer running at NASAs Ames Research Center, in Mountain View, Calif., achieved a sustained peak of 42.7 teraflops. The supercomputer is built from SGIs Altix systems and powered by Intel Corp.s 64-bit Itanium processors, according to officials with SGI, also in Mountain View.

When completed, Columbia will comprise 20 Altix systems. The 42.7-teraflop mark was reached using 16 of those systems powered by 10,240 Itanium processors.

Such massive single-box supercomputers are useful for particular niche applications, such as modeling for weather-forecasting jobs.

But while there can be some trickle-down benefits for enterprise IT in the research, it doesnt have the same impact as work on cluster computing, said Gordon Haff, an analyst with Illuminata Inc.

"To the degree that high-performance computing is relevant to the enterprise—and it increasingly is—its more about cluster technology," said Haff in Nashua, N.H.

/zimages/1/28571.gifCheck out eWEEK.coms Infrastructure Center for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.