On Tue, 16 Oct 2001, Andrew Sharp wrote:
> > > For instance, some time ago I asked what x86 (mobile)
> > > processor the iBook2's could be compared to and I received all
> > > answers from a Celeron 300 to a Pentium III 800. :-/ What is
> > > a realistic figure?
>
> Is this a troll?
I didn't take it for a troll - in this list, I usually don't feed trolls
(Branden and Ethan being the exception to this rule :->)
> > Irrelevant. Any figure someone could come up with is pretty irrelevant.
>
> There is not much hard data here. I've done considerable
> professional work in benchmarking and performance analysis, and the
No pun intended - there's just lies, damn lies, statistics and benchmarks,
in roughly that order. Having a fair benchmark that gives a meaningful
comparison of different processor architectures is tough. And anything
else is benchmarketing ;-)
> problem here is that Motorola doesn't submit the G3 or G4 processors
> for SPECmark testing, so it is damn difficult to compare it to other
> processors on an even basis. For example, Apple has published BS
There gotta be a reason for that :-) But even having these figures
wouldn't necessarily tell you enough to predict how a particular
application compares on the different processors, no?
> figures comparing the G4 to Pentium II processors several years
> ago. The problem is that these were vaguely presented, and were
> instruction mix specific. So, like, they picked a narrow mix of
> integer intructions that the G4 kicked butt on, and they came out
> smelling like a rose. The funny part was that the P-II actually
> beat the G4 on one of the categories, and Apple actually published
> that too in their ads. Just shows you that Apple is stupid, even
> when they are being stupid.
Apparently they're honest about being stupid. Pretty stupid :-)
> As for the iBook2, I would be surprised if it couldn't play DVDs.
Sure it can - with the hardware decoder used (assuming it has such a
beast; the Lombard does). Figure out how that's used by MockOS, and write
a driver.
I imagine playing DVDs pretty much clashes with preemptive scheduling and
other tasks using the CPU under Linux (if the CPU is maxed out by the
decoder).
> > I've seen anything from a bit worse than the clock speed ratio (blame that
> > on compiler optimization differences) to quite a bit better than the clock
> > ratio (where cache size got limiting on the mobile x86). That was with a
> > rather integer heavy app (but by no means exclusively integer math).
>
> My gut feeling is that the G3/G4 core is about 20% faster in a
> typical mix of non-GUI Linux applications, so a 500MHz G3 is
> probably loosely comparable to a P-3 600MHz, but with *much* better
> power consumption. I've said it before, I personally think the
I came up about even with number-crunching tasks (restrained molecular
dynamics and the like), and that's perhaps mostly due to better
optimization on Intel. Your 20% for other apps sounds reasonable.
Power consumption and heat didn't even factor in for me - though it's
striking how quiet the G3 runs compared to a PIII laptop that has the fan
kicking in half of the time. Plus it gets the same battery lifetime as the
PIII running at half speed in low power mode IIRC.
> > > I also understand that GCC may not be as mature for PowerPC as
> > > it is for x86, but is the difference considerable?
> >
> > Yes.
>
> Really? What would be an example of a good compiler for powerpc? I
Is there any (free one) ?
> wasn't aware that gcc was that bad, but I wouldn't know, I guess. A
> better question might be, in what areas does gcc suck in powerpc
> compilation?
Dunno - the consensus on linuxppc-dev seemed to be there's been more work
going into the intel version WRT optimization so it performs a lot better.
Other than that, just a gut feeling (I would have expected better
performance in my test cases).
Michael
