IBM’s BlueGene/L system installed at Lawrence Livermore National Laboratory (LLNL) ranked as the fastest supercomputer in the world at 280.6 TFlop/s, twice the performance the system demonstrated in earlier rankings. It is the first system to surpass the 100 TFlop/s mark and is expected to remain at the top of the rankings for some time.
IBM has been building out the custom cluster architecture at LLNL. Ultimately it will sport 128,000 custom Power CPUs when it is completed. BlueGene/L is designed to use less power and require less floor space than competing supercomputers, in part by its use of very small individual nodes with only a few custom-built chips.
Another version of BlueGene/L being built at IBM’s T.J. Watson Research Center ranked as the second most powerful computer in the world at 91.20 TFlop/s Linpack performance. A separate IBM architecture at LLNL, dubbed ASIC Purple and based on IBM’s pSeries 575 servers, took third place on the list at 63.4 TFlop/s.
The Columbia system at NASA/Ames, a cluster based on Intel Itanium CPUs built by SGI, slipped to the No. 4 spot with 51.87 TFlop/s performance.
IBM continues to be the dominant vendor on the list with 43.8 percent of the systems, slightly up from 43.2 percent a year ago. Hewlett-Packard maintained its second-place standing in the survey with 33.8 percent of all systems.
IBM’s systems represented 52.8 percent of the total installed performance of computers on the list, compared to 18.8 percent for HP. No other computer maker captured more than seven percent in any category.
Intel microprocessors are at the heart of two-thirds (333) of all 500 systems. Eighty-one of the systems used Intel’s latest 64-bit Xeon processors. AMD’s 64-bit Opteron processors gained ground with 55 systems using them compared to only 25 systems in the ranking six months ago.
IBM had the second most popular processor architecture among top 500 systems. Its Power chips appeared in 73 systems.
Clusters continue to be the most popular architecture among systems on the list with 360 clusters in the current ranking. That’s up from 296 clusters a year ago. On the current list, 249 clusters used Gigabit Ethernet as an interconnect, while 70 used Myricom’s Myrinet.
The least powerful system on the list was measured at 1.64 TFlop/s. That’s up from 850.6 GFlop/s for the lowest ranking system one year ago.
The Top 500 list is compiled by Hans Meuer of the University of Mannheim, Germany; Erich Strohmaier and Horst Simon of Lawrence Berkeley National Laboratory; and Jack Dongarra of the University of Tennessee.