PPC 970 = high performance?

http://www.theinquirer.net/?article=10180

http://www.sandpile.org/impl/pics/am...013_1m_ovr.jpg

Above are URLs for die photos of the PPC970 and Opteron dies, both
..13u, 58 million transistors for the PPC and over a hundred million
for the Opteron.

The easiest way to assure high performance in a modern CPU is to use
large caches. The Opteron die uses 60% or more of the die area for
caches.

The easiest way to cheapen a CPU is to use just a teensy little bit of
cache. The PPC970 uses only about 15% of the die area for cache.

And yet we have to read all the PR drivel about Apple's G5 being a
high performance system! Well, Apple wants a high profit margin
(hence a really cheap CPU) and IBM wants to protect its Power CPUs
from cheap competition (hence a really slow CPU with a busted cache
system).

G5 = high performace? Bad joke.

On Thu, 26 Jun 2003 15:51:48 GMT, "Felger Carbon"
wrote:


And yet we have to read all the PR drivel about Apple's G5 being a
high performance system!

Manufacturer-designed, proprietary SPEC=invitation to cheat and
mislead.

RM

They add an external L3 cache, as they did with G4s. You are only
partly right: IBM is to blame, Apple is not.

Furthermore, software for Apple PCs tends to be more optimized that
x86 PC software. Typical optimizations are particularily important
when it comes to cache and memory usage. Apple has a good microkernel
Unix OS, x86 PCs... well.

Uli

--
------- http://grassomusic.de -------

Manufacturer-designed, proprietary SPEC=invitation to cheat and
mislead.

RM

Wouldn't that pretty much describe Intel's compiler?

I mean it was designed from the ground up and hand tuned to specifically get
the best possible results on benchmark tests.

gcc is pretty decent on x86. It sucks rocks on PowerPC. I'm amazed that it even
turned in a decent result on the 970.

IBM's AIX compilers do a far better job on Power and PowerPC architecture than
Apple's pathetic choice of gcc. I've never understood why Apple continues to
use gcc either. It stinks.

In article ,
nuke wrote:
Manufacturer-designed, proprietary SPEC=invitation to cheat and
mislead.

RM

Wouldn't that pretty much describe Intel's compiler?

I mean it was designed from the ground up and hand tuned to specifically get
the best possible results on benchmark tests.

gcc is pretty decent on x86. It sucks rocks on PowerPC. I'm amazed that it even
turned in a decent result on the 970.

IBM's AIX compilers do a far better job on Power and PowerPC architecture than
Apple's pathetic choice of gcc. I've never understood why Apple continues to
use gcc either. It stinks.


It's common between the two compared platforms. One amusing thing,
complaints were made about hyperthreading being turned off, Apple said
they did a compile with hyperthreading enabled and it gave slower results.
I think they did a fine job of refuting any whines.

http://apple.slashdot.org/article.pl...id=126&tid=181