Slashdot Mirror
Award of $200M Supercomputer To IBM Proving Controversial
An anonymous reader writes "According to documents accidentally placed on a federal government Web site for a short time last week the national science foundation (NSF) will award the contract to buy a $200M supercomputer in 2011 to IBM. The machine is designed to perform scientific calculations at sustained speed of 1 petaflop. The award is already proving controversial however, with questions being raised about the correctness of the bidding procedure. Similar concerns have also been raised about the award of a smaller machine to Oak Ridge national lab, which is a Department of Energy laboratory, not a site one would expect to house an NSF machine."
Most of the major research labs in the US are technically owned by the DOE, whether they're primarily weapons/classified research (Livermore, Los Alamos) or closely linked to academia (Berkeley, Fermilab).
The DOE and the NSF fund various projects (with some subject area overlap) but it's still up to individual scientists to write proposals asking for supercomputer time.
IBM has at least 3 different systems that this could be (x86-ish linux cluster, power* aix, bluegene), but the article doesn't say which.
Other competitors would have been Sun (linux or solaris), SGI (Altix), Cray, etc. Apparently the USGov won't consider Japanese machines so Hitachi/NEC are out.
PS Japan is building a 10x faster machine in the same timeframe.
PPS The top500 is heavily biased toward clustery machines; certain types of science codes do not run well on such systems. Not saying it's a wrong metric, just that the machines on the top500 are not good at every kind of problem.