Here is an interesting message which was posted by Michael Turner (leap@gol.com) to the main discussion forum. When will we see the first PS/2 cluster running a CFD code? At $200 per CPU it would be quite cost effective:

==

Computational Chemistry on a PS/2:

http://spawn.scs.uiuc.edu/research/sonyps2/ps2project.htm

The authors point out that the loss-leader model for marketing game systems (take a loss on the console, make money off the games), plus the economies of scale in that market, favors the use of video game boxes for COTS clustering.

Their usability as general purpose computers apparently owes something to EU trade policy, amusingly enough:

http://www.theregister.co.uk/content/2/14534.html

The note concludes: "It is fortunate for computational chemistry that many of the operations which need to be done quickly for realistic games are essentially the same as the ones which form the bottleneck in quantum chemistry. We expect that once the COTTS [sic] model is demonstrated to be effective for computational chemistry, other scientific computing applications will follow, and the idea could have a widespread impact similar to the Beowulf cluster."

It's been donkey's years since I've done anything in computational chemistry, so I'm not sure if this statement applies, mutatis mutandis, to computational fluid dynamics. The authors suggest that their raw performance numbers for dot-product in assembly code (not carefully optimized) probably understate what this box is capable of doing.

Michael Turner

• Sony PS/2 Clustering - News Editor, Tue, 30 Jul 2002, 11:15 a.m.
  • Re: Sony PS/2 Clustering - Robin, Tue, 6 Aug 2002, 3:56 p.m.