Two systems making their first appearance on the Top 500 list of the world’s fastest computers underscore the broad range of today’s high-performance computing options: A self-made cluster of 1,100 Apple G5s at Virginia Tech landed the No. 3 spot in the list, highlighting the recent trend toward do-it-yourself clustered systems built with low-cost, off-the-shelf components (not to mention Apple’s debut in the biannual ranking; see story, p. 1). Meanwhile, a prototype of IBM’s BlueGene/L — a highly customized system using a new architecture under development at the IT giant’s Watson research labs — entered the list at the No. 73 spot.
While the vector-processor-based 35.9 Tflop/s Earth Simulator still topped the 22nd edition of the list, released Nov. 16, cluster-based systems surged ahead, with seven clusters in the top ten and a total of 208 systems — up from 149 six months ago. For the first time, clustered systems are the most common computer architecture on the list.
Commodity chips are also gaining ground. The number of systems using Intel processors grew in the last six months from 119 to 189, making Intel the top processor family used in high-performance systems, followed by Hewlett-Packard’s PA-RISC chips and IBM’s Power architecture.
The BlueGene/L prototype was the only new life science computer on the list. The top bioinformatics supercomputer remains the 4,000-processor Dell cluster at the Buffalo Center of Excellence in Bioinformatics, which slipped from No. 25 to No. 38. A revised benchmark score bumped the 1,048-processor IBM system installed at the Translational Genomics Research Center up from No. 66 to No. 53. Three systems on the June 2003 ranking — at the Stanford University Biomedical Computational Facility, the University of California Santa Cruz, and Bayer — did not meet the current list’s lower limit of 403.4 Gflop/s.
IBM still leads the Top 500 list in terms of overall performance, with a total of 188 Tflop/s. HP is next, with a combined performance of 119 Tflop/s, followed by NEC with 46 Tflop/s. HP still boasts the greatest number of installed systems on the list, however, with 165 — up from 159 in June. IBM follows with 159 systems, and SGI is next with 41.
The Top500 list now includes 131 systems that report a Linpack performance over 1 Tflop/s, and the total combined performance of all 500 computers on the list was 528 Tflop/s — more than half a petaflop — compared to 375 Tflop/s six months ago.
The Top 500 list is compiled by Hans Meuer of the University of Mannheim, Erich Strohmaier and Horst Simon of Lawrence Berkeley National Laboratory, and Jack Dongarra of the University of Tennessee. The complete list is available at http://www.top500.org/.