Supercomputing performance at 1/10th the cost

[Greg Lindahl]

> I've read in several explanations of
> Beowulf clusters that they can provide equivalent super computing
> performance for 1/10th the cost.  I believe this to be true, but
> would like to back it up with some example cases.

Most of the examples I've seen assume that the "super computer" is obsolete
or being stupidly used.

Even if you want to compare to modern machines (IBM SP2), it can be hard to
figure out pricing, because most announced deals are multi-year deals with
lots of phases and nodes with memory and storage and etc. Sure, you could
use list prices, but who pays list?

-- g