decent performance from G4 Macs?

[Mark Hahn]

> --- Mark Hahn <hahn at physics.mcmaster.ca> wrote:
> > I'm doing some benchmarks to evaluate whether
> > current Macs would make suitable nodes for a serial
> > farm (lots of nodes, preferably fast CPU and dram,
> > but no serious interconnect.)
> 
> Physics or bioscience code?

why does it matter?  we're not trying specifically to run BLAST, 
if that's what you're asking.  I don't see any reason why the 
department would matter, but it's a mixture of math, chem,
physics, astro, biologists, and perhaps a few psychologists.

> > I've tried a variety of real codes and benchmarks,
> > but can't seem to get something like a Mac G4/800 
> > with PC133 to perform anywhere close to even a
> > P4/1.7/i845/PC133.
> > 
> > I'm using either the gcc 2.95 that comes with OSX or
> > a recent 3.1 snapshot (which is MUCH better, but
> > still bad).
> 
> What compiler are you using for the P4? 

I'm pretty happy with recent snapshots of gcc 3.1 (pre-release).
(still mystified why gnu fortran people are stuck at F77, but...)

> > is it just that the performance Apple brags about is
> > strictly in-cache, and/or when doing something ah
> > specialized like single-precision SIMD
> >(altivec/velocity engine)?
> 
> Apple has some libraries that take advantage of the
> Altivec instructions.

linpack/lapack/atlas/fftw?

> AFAIK, there are several people using MacOS X in
> clusters, the SGE (Sun Grid Engine) project has a port
> for Mac OS X.

which doesn't give me ANY data on performance.