Once again, IBM can lay claim to having the worlds fastest supercomputer.
The list of the top 500 supercomputers in the world was released June 27 at the International Supercomputer Conference in Dresden, Germany, with IBM taking top honors for the fourth straight time.
Big Blue, of Armonk, N.Y., built six of the top 10 supercomputers on this years list. Dell, Cray and Silicon Graphics also were included among the top 10. Unlike last year, when only one of IBMs Blue Gene/L systems cracked 100 teraflops of performance, at least three of the top 10 supercomputers in the 2007 list sped past the 100 teraflop mark.
The top 500 list, which is published twice a year, is compiled by the University of Mannheim, in Germany; the University of Tennessee; and the National Energy Research Scientific Computing Centers Lawrence-Berkeley National Laboratory.
“The 29th edition of the closely watched TOP500 list of the worlds fastest supercomputers shows a lot of shuffling among the top-ranked systems and the largest turnover among list entries in the history of the TOP500 project,” according to a statement from the three institutions that compile the list.
The worlds fastest supercomputer remained IBMs Blue Gene/L system installed at the Department of Energys Lawrence Livermore National Laboratory, in Livermore, Calif. The systems peak performance on a Linpack benchmark was still clocked at 280.6 teraflops, or 280.6 trillion calculations per second.
Unlike past lists, two other systems, both built by Cray, broke past 100 teraflops of performance. The upgraded Cray XT4/XT3 ranked second on the list, with a benchmark performance of 101.7 teraflops. That system is installed at the DOEs Oak Ridge National Laboratory in Tennessee.
Cray also built the Red Storm system, which is installed at the Sandia National Laboratory in Albuquerque, N.M., and ranks third with a peak performance of 101.4 teraflops.
The notion of supercomputers breaking the teraflop barrier might seem obsolete soon. Both IBM and Sun Microsystems said at the show that they are building or will build systems that offer a “petaflop”—or a thousand trillion calculations per second—of performance.
All of the systems on the top 500 list are getting faster. The entry level mark for the list increased from about 2.74 teraflops six months ago to about 4 teraflops now.
Since 2004, IBM has been ahead of Hewlett-Packard in terms of the number of systems represented on the list. However, IBMs share now stands at 38.4 percent, compared to 47.2 percent a year ago, while HPs share increased from 31.6 percent to 40.6 percent.
The explosion of x86 microprocessor architecture continues within the top 500 list. A total of 289 systems on the list use Intel processors – a combination of Xeon and Itanium processors – and 205 of those systems use the chip makers dual-core Xeon processors. While the number of systems using Advanced Micro Devices systems dropped, the company still powers 105 systems on the top 500 list.
AMDs Opteron chips have now passed IBMs Power processors, which are used in 85 systems.
Of the top 500 supercomputers, 373 systems are labeled as clusters, which makes this the dominant platform architecture on the list.
Of the top 10 supercomputers, eight are installed at various facilities in the United States. The fastest supercomputer in Europe is an IBM JS21 clusters, which offers 62.6 teraflops of performance and is ranked ninth on the list. This system is installed at the Barcelona Supercomputing Center in Spain.
In Japan, the highest-ranking system is at the Tokyo Institute of Technology, which comes in as number 14 on the top 500 list. This cluster is integrated by NEC and uses Sun Fire x4600 servers with Opteron processors and an InfiniBand interconnect.