IBM again tops supercomputer list

Moore’s Law appears to be holding its own, even at the most crushing levels of systems performance — but even in the world of supercomputers, the performance ramp appears to be bending under the strain of higher frequencies. The reigning three-time champion, IBM’s BlueGene/L System installed at the DoE at the Lawrence Livermore National Labs, once again sits atop the list of big iron, recording a Linpack benchmark performance of 280.6 TFlop/sec (i.e., trillions of calculations per second), remaining the only system ever to exceed the 100 TFlop/sec level.

The top 10 of this year’s list saw some changes, with three systems removed and three added. The largest system in Europe entered the list at No.5, an Itanium-based system built for the Commissariat a l’Energie Atomique (CEA) in France. Japan’s largest system, a Sun and NEC Opteron-based cluster at the Tokyo Institute of Technology, gained the No. 7 spot, while another IBM BlueGene system at the German Forschungszentrum Juelich (FZJ) achieved the No. 8 position. The “Earth Simulator,” which topped the list for five years before being usurped by BlueGene/L in late 2004, has slipped to No. 10.

While significant turnover in the top-500 supercomputer list is normal, this year saw less than usual, as only 158 systems from a year ago failed to make the grade this year. Still, average performance continues to be stable and keep pace ahead of Moore’s Law, according to the organizers of the 27th annual rankings: Hans Meuer of the U. of Mannheim/Germany; Erich Strohmaier and Horst Simon of NERSC/Lawrence Berkeley National Laboratory; and Jack Dongarra of the U. of Tennessee/Knoxville.

Intel-based systems (IA-32, EM64T) accounted for more than half (53%) of all the top 500 supercomputers, while 81 systems (16%) have AMD processors at their core, vs. just 25 (5%) a year ago. Based on performance, roughly half of all systems used IBM as the primary architecture (48.6% of systems, 54.3% of installed performance), followed by HP (30.8%, 17.5%).

Other interesting statistics: The baseline performance mark to be considered for the list nearly doubled in 2006 to 2.03 TFlop/sec, vs. 1.17 a year ago, while the last system on this year’s list was ranked No.183 in June 2005. The entry level for this year’s top 10 exceeded 35 TFlop/sec, with the entry point to the top 100 increasing to 4.71 TFlop/sec. Total combined performance of all 500 systems in the rankings is a mindbending 2.79 petaflops/sec (a thousand teraflops), vs. 1.69 PFlop/sec a year ago.

For simpler perspective on just how powerful computers have become — the most recent Top 500 supercomputer list has nearly 10x the horsepower of the total accumulated performance from November 2002 — a number that has now been roughly matched by BlueGene/L alone.

The full list of the top 500 supercomputers, with lists and breakdowns by category including performance, architecture, and region, can be found at


1 DOE/NNSA/LLNL (US) BlueGene/L – eServer Blue Gene Solution (IBM) 131072 2005 280600 367000
2 IBM Thomas J. Watson Research Center (US) BGW – eServer Blue Gene Solution (IBM) 40960 2005 91290 114688
3 DOE/NNSA/LLNL (US) ASC Purple – eServer pSeries p5 575 1.9 GHz (IBM) 12208 2006 75760 92781
4 NASA/Ames Research Center/NAS (US) Columbia – SGI Altix 1.5 GHz, Voltaire Infiniband (SGI) 10160
2004 51870 60960
5 Commissariat a l’Energie Atomique (CEA) (France) Tera-10 – NovaScale 5160, Itanium2 1.6 GHz, Quadrics (Bull SA) 8704 2006 42900 55705.6
6 Sandia National Laboratories (US) Thunderbird – PowerEdge 1850, 3.6 GHz, Infiniband (Dell) 9024 2006 38270 64972.8
GSIC Center, Tokyo Institute of Technology (Japan) TSUBAME Grid Cluster – Sun Fire X64 Cluster, Opteron 2.4/2.6 GHz, Infiniband (NEC/Sun) 10368 2006 38180 49868.8
8 Forschungszentrum Juelich (FZJ) (Germany) JUBL – eServer Blue Gene Solution (IBM) 16384 2006 37330 45875
9 Sandia National Laboratories (US) Red Storm Cray XT3, 2.0 GHz (Cray) 10880 2005 36190 43520
10 The Earth Simulator Center (Japan) Earth-Simulator (NEC) 5120 2002 35860 40960


Easily post a comment below using your Linkedin, Twitter, Google or Facebook account. Comments won't automatically be posted to your social media accounts unless you select to share.