Another point is that, with the probable exception of embarrassingly parallel codes (eg, ray trace), adding infinite hardware is currently a non-solution. If using MPI to parallelize the code, the practical limit is probably 128 nodes, possibly up to 256 or so with heroic effort. If using OpenMP, you are limited to 32 CPUs on OS X (typical of most UNIX systems; the limit under Windows is 2-4 CPUs, depending which version).
To me, the bigger issue, because I need to write code that is portable across multiple systems, is the cost to incorporate Altivec since it is a single system solution. For some codes, eg, Craig Hunter's Jet3D, Altivec is definitely a winning solution and worth the cost. But that is not the case for all computational codes. As the old saw goes, your mileage may vary.
Don Jones
"Easy things should be easy and hard things should be possible."
-- Larry Wall
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Scitech mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/scitech/email@hidden
This email sent to email@hidden
