Friday, May 9, 2008

LBNL "ipod supercomputer"

http://uk.news.yahoo.com/vdunet/20080507/ttc-us-boffins-design-ipod-supercomputer-6315470.html

M. Wehner, L. Oliker, J. Shalf, "Towards Ultra-High Resolution Models of Climate and Weather", Internation Journal of High Performance Computing Applications (IJHPCA), April, 2008.

This system proposes a custom CPU core, which makes sense given the large scale design. Its somewhat similar to the SiCortex machine in many ways.

1 comment:

Platypus said...

Yes, it does look a lot like the SiCortex machines, and I work at SiCortex so I think I can say that (though I don't speak for them yadda yadda). If they actually try to build such a system, they'll run into many of the same issues as we have, from mechanical and power-distribution stuff up to system-management issues. In a way, compute cycles are the easiest problem here. Getting the data around - from external storage into the system, from node to node, from processor to memory, through whatever caches are involved - is the tough one. They'd also run into problems we haven't - e.g. would it be physically possible to arrange the pins and traces and cables they'd need for the required degree-6 diameter-8 Kautz network or equivalent? - but the result would probably still end up looking a lot like an SC5832 or BG/L in many important respects.

I sure wish we'd come up with "iPod supercomputer" though. ;)