In article <1133373979.773585.275720@z14g2000cwz.googlegroups .com>, "Mark
Taylor" <mky@talk21.com> writes:
> >> should it make a difference between P4 and P5 ?
> >> I'm running 5.2 "off-the-shelf" (CD)
>
> 5.2 doesn't support smt.

So should I expect significantly better performance
with 5.3/smt ?

>
> >> What's the big difference here, if one just compares single CPU performance ?
>
> you didn't state this in your original post
> >> Shouldn't one expect a bit more "intrinsic" gain when
> >> going from P4 to P5 ?
>
> That would wholly depend on your application, what were you expecting ?

Well, I don't think my app(s) are so very special, some double precision float,
some integer, addressing, branching. No Linpack type stuff, however.

I was expecting at least some performance gain, let's say
around 1.5 .

When I compare the very same code on e.g. an ancient F50 ( 333 MHz PPC 604e )
vs a less ancient 44P ( 375 MHz Power3 ) I measure a speed gain of
1.5, which is well above the clock ratio.
Comparing that Power3 box with a p630 or p615 ( 1.2 Ghz Power4 )
I see a gain of 2.4, which is well below the clock ratio, but since
there's a net gain I wouldn't complain.
Now comparing that Power4 box with a shiny new p520 ( 1.5GHz Power5 )
the gain is a less-than-stellar 15%. From clock ratio alone I would
have expected 25%.
My other p615 (1.4 GHz Power4) even beats the p520 by 5% !
So where's the incentive to go Power5 ?
And yes, I've tried different compiler versions as well as tuning options,
the differences are always negligible.
Which is consistent what I've seen in the past, all my codes seem
to be resistant against -qtune & friends :-(