[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Stata (gllamm) benchmarks for different platforms?
> Chinh Nguyen:
> Our belief that any speed improvement from optimized code would be minimal was
> based on the information we had at the time. In addition, an Apple engineer
> personally told me that breaking the 4GB limit was very difficult in OS X at
> that time (late 2003, when Panther was available) without inside knowledge and
> help from Apple.
I'm happy to hear that you keep on bothering Apple to get the most out of
the system. We have a Swedish saying: The dog on the cat, the cat on the
mouse, the mouse ... (I have no idea if this makes any sense in English)
But, maybe you should also be knocking on IBMs door. They have a new
(January 16) XL compiler that features: "Generation of highly optimized code
which exploits the IBM PowerPC architecture within the Power Mac G5" and
"Binary compatibility and coexistence with gcc V3.3".
The (amateur) benchmarks I have seen (for an earlier version of XL) are very
impressive and suggests that the gcc and the XL doesn't even play in the
same league. If you are really serious about producing high performance
code, the IBM XL compiler seems to be the obvious choice. And this works for
other Macs as well (at least the G4 and the G5).
> However, it's
> doubtful you're going to see the 3 to 4 times improvement you're hoping for
> when we go 64-bit.
Unfortunately, you are probably right. I still though, have a hard time
accepting the G5 to have only about 27% of the Opteron performance. If so,
the guys at Virginia tech really made a poor choice building a supercomputer
on the G5 instead of the Opteron. And they still managed to get to third
place in the world.
It would be interesting if you could also test a 32-bit Stata version on the
Opteron. That would give a rough idea of how much speed improvement one
could expect to "go 64-bit".
* For searches and help try: