IBM overtakes HP in top of the teraflops
Supercomputer hit parade...
IBM has replaced HP as the hostest with the mostest fastest supercomputers. In the 23rd bi-annual survey of Top 500 supercomputer sites, IBM has 44.8 per cent of entrants and 50.12 per cent of the performance. HP is in second place with 28 per cent of systems and 18.5 per cent of installed performance.
Pole position is held for the third year in a row by NEC's Earth Simulator based in Yokohama, Japan. It has a Linpack benchmark performance of 35.86 teraflops - trillions of calculations - per second.
In second place is Thunder at the University of California's Lawrence Livermore National Laboratory which uses more than 4,000 Intel Itanium 2 chips. It supports the university's research activities in materials research, electromagnetics, atmospheric and ocean science.
Missing this time around was Virginia Polytechnic Institute and University's machine based on 1,100 dual G5 processors. It dropped out of the charts, from third place, as its hardware was being upgraded at the time of the compilation of the survey.
Clustered systems account for the majority of entrants with 291 machines, up from 208 in November 2003. Intel chips are at the heart of 287 systems, up from 119 last year. IBM Power Processors are used in 75 systems, HP PA-RISC chips are in 57 and AMD chips in 34 systems.
The UK outranked Germany for the first time with 36 supercomputers in the top 500 compared with 34 in Germany.
Japan is a traditional powerhouse for supercomputers, but other Asian countries are also increasing their presence. China made it into the top ten for the first time at number ten. The Dawning 4000A is based on 2,000 AMD Opteron chips and is used by the Chinese Academy of Sciences' Supercomputer Center in Shanghai.
Asian countries, excluding Japan, has 55 supercomputers in the Top 500. Japan has 34 entries and Europe has 124.
The Top 500 list is compiled twice a year by academics for the International Supercomputer Conference. ®
Sponsored: Benefits from the lessons learned in HPC