Heading the list is IBM’s BlueGene/L system, with an eye popping 70+ teraflops. That’s over 70 trillion floating point operations per second. IBM expects the completed BlueGene/L machine to be capable of 360 teraflops. IBM is already selling the system to interested facilities at a starting pricetag of $1.5 million.
Number two on the Top 500 is NASA’s Columbia Supercomputer. Columbia is comprised of 10,240 clustered Itanium 2 processors on SGI Altix servers, while IBM’s behemoth features 32,768 PowerPC 4 processors. IBM’s machine is something of a throwback, being a computer built explicitly for supercomputing purposes – a trait it shares with NEC’s former #1 (and current #3) supercomputer, the Earth Simulator. Most of the Top 500 supercomputers are actually immense clusters of high end, commercially available servers, linked via high speed Ethernet and special clustering software available for Unix or Linux systems.
Coming in at number 7 on the supercomputer list is the System X supercomputer from Virginia Tech. System X is constructed of 1,100 Apple XServe G5 blade servers. SystemX actually ranked #3 last year and the slip to #7 is a bit disingenuous. SystemX actually increased its performance by 20% over the last year. Only the appearance of several new systems like BlueGene/L and Columbia caused the drop in the machines rankings. And SystemX still puts up some impressive numbers, with the Mac OS X cluster pumping out 12.25 teraflops. The speed with which these massive commodity clusters are appearing is incredible. The technological barriers to supercomputing are dropping all the time. Maybe I’ll cook one up with a couple dozen PlayStations and see what happens.