SCIENCE

When the 28th edition of the closely watched TOP500 Supercomputers list is officially unveiled Nov. 14, the ranking of the top 10 sites will clearly illustrate how the field remains both constant and constantly changing. The latest edition of the list, which is published every June and November, will be presented and discussed in depth at a Birds-of-a-Feather session to be held at 5:30 p.m. Tuesday, Nov. 14, at SC06, the international conference of high performance computing and networking. SC06 will be held Nov. 11-17 in Tampa, Fla. The new TOP500 list, as well as the previous 27 lists, can be found on the Web at its Web site. Highlights from the latest list On the new list, the IBM BlueGene/L system, installed at DOE’s Lawrence Livermore National Laboratory (LLNL), retains the No. 1 spot with a Linpack performance of 280.6 teraflop/s (trillions of calculations per second, or Tflop/s). The new No. 2 systems is Sandia National Laboratories’ Cray Red Storm supercomputer, only the second system ever to be recorded to exceed the 100 Tflops/s mark with 101.4 Tflops/s. The initial Red Storm system was ranked No. 9 in the last listing. Slipping to No. 3 from No. 2 last June is the IBM eServer Blue Gene Solution system, installed at IBM’s Thomas Watson Research Center with 91.20 Tflops/s Linpack performance. The new No. 5 is the largest system in Europe, an IBM JS21 cluster installed at the Barcelona Supercomputing Center. The system reached 62.63 Tflop/s. Sandia’s Dell PowerEdge system was re-measured at 53 Tflop/s and was able to hold on to No. 6 spot. The NovaScale 5160 system built by the French company Bull and installed at France’s Commissariat a l'Energie Atomique (CEA) slipped to No. 7, despite a new Linpack measurement of 52.84 Tflop/s. The original system debuted in the No. 5 position in June. No. 9 is now occupied by the largest system in Japan, a cluster integrated by NEC based on Sun Fire X4600 with Opteron processors, ClearSpeed accelerators, and an InfiniBand interconnect. The system is installed at the Tokyo Institute of Technology. The current No. 10 system is the upgraded Cray XT3 system at DOE’s Oak Ridge National Laboratory with 43.48 Tflop/s. One year ago its predecessor was also No. 10 with just 20.53 Tflop/s. The Earth Simulator, built by NEC, which held the No. 1 spot for five lists, has now slipped out of the TOP10 and is ranked at No. 14. While the TOP500 list provides top systems with immediate bragging rights, the list also provides a useful overview of recent changes in the field of high performance computing hardware. Taken together over the years since its debut, the list is the only method which uses a consistent benchmark to measure system performance across various computer architectures. Overall trends and changes in the list - The entry level to the list moved up to the 2.737 Tflop/s mark on the Linpack benchmark, compared to 2.026 Tflop/s six months ago. - The entry point for the top 100 increased in six months from 4.71 Tflops/s to 6.65 Tflop/s. - The last system — No. 500 — on the latest list would have been listed at position 359 in the last TOP500 just six months ago. This is one of the smallest turnover rates seen in the history of the TOP500. - Total combined performance of all 500 systems on the list has grown to 3.54 petaflop/s, compared to 2.79 PFlop/s six months ago and 2.30 PFlop/s one year ago. Other highlights A total of 261 systems (52.2 percent) are now using Intel processors, down from 333 (66.6 percent) one year ago. The AMD Opteron family passed the IBM Power processors and is now the second most common processor family with 113 systems (22.6 percent, up from 55 system (11 percent) one year ago. Ninety-three systems (18.6 percent) use IBM Power processors, up from 73 systems (14.6 percent) one year ago. Dual core processors are becoming widespread. Already, 75 systems use Opteron dual core processors and an impressive 31 systems use the new Intel Woodcrest dual core chips. Gigabit Ethernet is still the most used internal system interconnect technology (211 systems, down from 256 six month ago). Myricom's Myrinet is used in 79 systems down from 87. InfiniBand technology was able to increase its share to 78 systems, up from 36 six month ago. At present, IBM and Hewlett-Packard sell the bulk of systems at all performance levels of the TOP500. IBM remains the clear leader in the TOP500 list with 47.8 percent of systems (unchanged from 48 percent) and 49.9 percent of installed performance (down from 53.9 percent). HP is second with 31.2 percent of systems (unchanged from 31.4 percent) and 16.5 percent of performance (down from 17.8 percent). The other manufacturers with more than five percent in any category are: Cray (8.2 percent of performance), Dell (7.3 percent of performance), and SGI (5.4 percent), each of which benefit from large systems in the TOP10. Of the 244 systems installed at commercial and industrial customers, IBM (121) and HP (116) together account for 237 systems, showing they have clearly cornered this important market segment. The U.S. is clearly the leading consumer of HPC systems with 306 of the 500 systems. The European share (95 systems) recovered slightly and is again larger then the Asian share (79 systems). In Europe, UK has established itself as the No. 1 with 32 systems (35 six months ago). Germany has to live with the No. 2 spot with 19 systems (18 six months ago). Dominant countries in Asia are Japan with 30 systems (up from 29) and China with 18 systems (down from 28).