On 5/29/11 11:28 PM, Sridhar Ayengar wrote:
Well, yes. A lot of early attempts at *anything* kind
of sucked. We
are now 20 odd years past the MPP "boom" of the mid 80s - I certainly
hope technology would have marched on during those years.
If your problem isn't embarrassingly parallel, you're doing the wrong
problem.
10 GOTO 10
Parallelize THAT.
Is the goal to decrease the execution time of one iteration, or perform
as many iterations as possible in one unit of time? ;)
Your question boils down to: Do we want the world's fastest NOP or the
most NOPs per second? :)
Peace... Sridhar