decent performance from G4 Macs?

[Robert Latham]

On Sat, Apr 13, 2002 at 02:18:39PM -0400, Mark Hahn wrote:
> 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)?

it's the altivec unit that makes G4s at all interesting.  if you
aren't using the vector unit, yeah, you won't even come close to x86.   

gcc is multi-platform, sure, but it's optimizer for x86 has received a
lot of attention, while the powerpc optimizer has not. your
observation that gcc 3.1 performance is better shows that focus on
powerpc optimizations has grown, but yeah, it's going to get less
attention than x86.  too bad, really.  register pressure on a powerpc
is much less than on x86 ( register pressure on just about any arch
not stack-based is less than that on x86 :> )

you are running on mac os x, yes?  is there any chance you could put
linux on it?  if your application is making a significant number of
system calls ( file i/o, network traffic... you know, system calls )
os x will hurt you. I'd be curious to hear if your application
performs better under linux on powerpc (debian, suse, mandrake,
yellowdog; there are many options) than it does under os x on the same
hardware.  ( if you use linux, you'll have to hand-code some assembly
to use the G4. samples abound on the web.  but if you are
compute-intensive anyway, you might not see gains running under linux)

microbenchmarks don't always correlate well with application
performance, but here are lmbench numbers. the hardware is constant
while i varied the operating system: 
http://clustermonkey.org/~laz/pbook/lmbench.powerbook.txt (the numbers
are nearly 8 months old, but the newer versions of os X do not show any
remarkable improvement and in fact regress on some scores)

rgb, do you know what the cputest curves look like for a G4 mac?  

also bear in mind that G4s run significantly cooler than their x86
counterparts, so you might still come out ahead on price/performance,
where price takes into account initial purchase + cost of running the
cluster.  

so there you go. there are lots of reasons why you'll have to actually
spend a bit of effort to move to a new architecture.  i hope no one on
this list finds that idea surprising.  

==rob

-- 
Rob Latham
                                             A215 0178 EA2D B059 8CDF  
                                             B29D F333 664A 4280 315B