May 12th, 2007, 4:21 pm
QuoteOriginally posted by: CuchulainnWhat kind of speedup would be needed/expected for quadcore?On a duo-core, our MC simulator was optimised 1) sequential 2) OpenMP. For the latter we got a speedup of 80% (1.8) which is almost superlinear. Will speedup on 4-core be 3.6?It's not a free lunch? At some stage you might need a combination of shared and distributed computing.It's pretty much a free lunch for MC, though. You are going to run into iterative algorithms that are inherently difficult to parallelize, but MC certainly isn't in this category. Hell, if you really wanted parallelism I'm sure you could do MC on a graphics card, and just resign the cpu to generating "random" numbers (you could do this on the graphics card too, but it would be difficult to implement safe generators, at least last I checked).