IBM Dominates Semiannual Top 500 Supercomputer List

The company's BlueGene/L system has climbed from 73rd to first place, and five of the top ten supercomputers are also from IBM.

For the first time, IBM swept the top three rankings and five of the top nine in the most recent listing of the worlds fastest supercomputers.

The 26th semiannual Top 500 Supercomputer Sites listing was released Monday on the independent Web site.

Big Blues defending champion, the Department of Energys IBM BlueGene/L system, installed at the University of California-Berkeleys Lawrence Livermore National Laboratory in Livermore, Calif., won easily over the competition, clocking processing speeds nearly three times as fast as those of the average human brain, the survey said. BlueGene/L uses 131,072 processors and is roughly the size of a big-screen television.

The BlueGene/L production team doubled the computer in size and capacity during the last six months to achieve a record Linpack Benchmark performance of 280.6 TFlop/s (teraflops)—the only system ever to exceed the 100 TFlop/s mark, the survey said.

The Linpack Benchmark is a math function that measures how fast a computer solves dense systems of linear equations.

Only two years ago, the BlueGene/L system was ranked 73rd on the Top 500 list.

IBM also took the honors for most computers on the Top 500 list with 219.

Two of the worlds top three supercomputers are located at the Livermore facility. The third-place computer, the IBM ASCI Purple eServer pSeries 575, with 10,240 processors, is also part of the DOEs computing lineup. It reached a speed of 63.4 TFlop/s.

/zimages/4/28571.gifClick here to read more about IBMs previous two winners in the supercomputing ranks.

The No. 2-ranked computer, a similar but smaller (40,960 processors) IBM eServer Blue Gene system installed at IBMs Thomas Watson Research Center in Yorktown Heights, N.Y., was far behind the overall winner in speed, recording a 91.20 TFlop/s Linpack performance.

The U.S. Department of Energy can claim ownership of the fastest, most powerful computing systems in the world, owning four of the worlds top six computers. In addition to the Nos. 1 and 3 computers in California, two systems at the DOEs Sandia National Laboratories were ranked Nos. 5 and 6: A new PowerEdge-based Dell system outperformed the enlarged ASCI Red Storm system by a narrow margin with 36.10 Tflops/s versus 35.86 Tflop/s to enter the top 10 for the first time.

Three of the worlds top four supercomputers are located in California. The Columbia system at NASA/Ames in Mountain View, Calif., built by SGI, slipped to the No. 4 spot from No. 3 last June, clocking a still-impressive 51.87 TFlop/s.

Besides the dominance of IBM and the DOE, perhaps the most noteworthy aspect of the November 2005 listing is that Japan has slowly but surely begun to fade in the supercomputing benchmark competition. After leading the survey for five consecutive lists—from June 2002 to June 2004—the Japanese-built NEC Earth Simulator Center computer has slipped to No. 7. The Earth Simulator placed third a year ago and was fourth last June.

Still, the Earth Simulator was the only non-U.S. system to crack the Top 10.

/zimages/4/28571.gifApple servers power a new EU supercomputer. Click here to read more.

Cray, the first name in supercomputing only a few years ago, took sixth and 10th place with Red Storm at Sandia National Laboratories and Cray XT3 at Oak Ridge National Laboratory, respectively.

Hewlett-Packard Co. was second to IBM with 169 systems on the overall list. Cray and SGI were third with 18, Dell had 17, and Linux Networx had 16. No other company ranked in double figures on the list.

The United States had 305, about 61 percent of the systems on the Top 500 list. The United Kingdom, with 41, and Germany, with 24, were second and third, respectively. Japan was fourth with 21 systems.

Also on Monday, IBM showed a preview of the next-generation 64-bit POWER5+ p5-575 supercomputer at the SC2005 Conference in Seattle, Wash. The new computer will allow up to 128 sixteen-processor p5-575 cluster nodes to create a single high-performance system with more than 2,000 CPUs and high-performance capabilities, IBM said.

Introduced just last month, POWER5+ is a "server on a chip" containing two processors, a high-bandwidth system switch, a large memory cache and an I/O interface.

POWER5+ p5-575 uses the same cluster building-block package to extend the power of the POWER5 p5-575—the Big Blue system at the core of supercomputers that power work in areas such as genome research, automotive crash-testing, petroleum exploration, and oceanographic, atmospheric and energy studies.

Built on IBMs 64-bit POWER5+ technology, the p5-575 is planned to be available with 1.9 GHz and 2.2 GHz POWER5+ processors and support for AIX 5L(TM) Version 5.2 and 5.3, as well as Linux operating systems.

The ECMWF (European Centre for Medium-Range Weather Forecasts) will be one of IBMs first customers for large clusters incorporating the new POWER5+ p5-575 systems with initial implementation planned during the first half of 2006, the company said.

/zimages/4/28571.gifCheck out eWEEK.coms for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.

Chris Preimesberger

Chris J. Preimesberger

Chris J. Preimesberger is Editor-in-Chief of eWEEK and responsible for all the publication's coverage. In his 13 years and more than 4,000 articles at eWEEK, he has distinguished himself in reporting...