Apple hardware (was: From a NY Times Bestseller)

[Michael Costolo]

> > I've personally seen Matlab code that took a day to run on a dual
> > Xeon Win 2000 machine complete in 10 minutes on a Powerbook G4.
>
>   1440 minutes reduced to 10 minutes?  That is 14400 percent.  I don't
> think you can attribute that to an increase in FSB alone.  :-)

No, not at all.  My point was more that the hardware was (or at least
seems to be) engineered to function (optimally?) as a system, rather
than just to plop whatever processor currently had the fastest clock
speed into the cheapest motherboard one can find and rush it out the
door.

Though approach #2 does seem to sell more computers...

>   Matlab is kind of weird.  (Like most software.)

AGREED!

>  I know it is
> single-threaded, so it runs at the same speed regardless of whether
> you have one, two, or 50 processors.  But even that wouldn't account
> for that difference.  Given the nature of typical Matlab work (many,
> many math ops), I'd guess it was a processor-specific optimization.
> Wasn't Apple touting some SIMD thing with the PowerPC for awhile?
> That *could* make a huge difference for that kind of thing.

That's entirely possible. It was all math, and a few hundred million
iterations.  I know that I've seen lots of other physicists running
Apple hardware, perhaps for that reason.  If you abuse mathematics for
a living, that level of math operation performance is worth the
premium you pay.