[Eug]

Just downloaded the latest CHUD and ran Skidmarks on my 1 GHz TiBook just for the hell of it: Int-95, FP-99, Altivec-98. When I ran it a few months back (older version of Skidmarks? I dunno) on X.2 I got 100 across the board, which makes sense since because Skidmarks at 100 is supposed to be normalized to the G4 1 GHz.

Not that this really matters for real world usage, but I'm wondering if this is more evidence to suggest that Panther is more specifically optimized for the G5.

I wonder what the score differences between X.2.8 and X.3 would be on the G5.

Does somebody have a good summary page of the performance enhancements of Panther? I've seen bits and pieces everywhere but no good summary page. And I don't mean the Apple PR material. I'd like something to read which is understandable by the layperson, but does get a little into the meat of things.

[Anand]

Originally posted by Eug:
Just downloaded the latest CHUD and ran Skidmarks on my 1 GHz TiBook just for the hell of it: Int-95, FP-99, Altivec-98. When I ran it a few months back (older version of Skidmarks? I dunno) on X.2 I got 100 across the board, which makes sense since because Skidmarks at 100 is supposed to be normalized to the G4 1 GHz.

Not that this really matters for real world usage, but I'm wondering if this is more evidence to suggest that Panther is more specifically optimized for the G5.

I wonder what the score differences between X.2.8 and X.3 would be on the G5.

Does somebody have a good summary page of the performance enhancements of Panther? I've seen bits and pieces everywhere but no good summary page. And I don't mean the Apple PR material. I'd like something to read which is understandable by the layperson, but does get a little into the meat of things.

Machine: 1.6 Ghz G5 with 1.25 GB Ram

In 10.2.8: 120, 223,161


In 7B85: 103,224,159


This is with Chud tools 3.01.

[Eug]

Hmmm... So the drop in Panther integer performance is there on both my TiBook and on the G5. I wonder why, and I wonder if it is represented on any real world apps.

Code:
10.2            Integer      Floats     Altivec
TiBook 1 GHz      100         100         100
G5 1.6            120         223         161

10.3            Integer      Floats     Altivec
TiBook 1 GHz       95          99          98
G5 1.6            108         224         159

By the way, I'm not concerned that the G5 1.6 doesn't score much higher than the TiBook 1 GHz in integer performance. G4 efficiency for integer IS supposed to be better, but we can't do a direct integer comparison across CPUs based on this bench alone. What concerns me more is the 10% drop on the G5 going from X.2 to X.3 (and 5% drop on the TiBook). But like I said, who knows what this means?

EDIT: Updated with new info.

[Anand]

Originally posted by Eug:
Hmmm... So the drop in Panther integer performance is there on both my TiBook and on the G5. I wonder why, and I wonder if it has any significance for real world apps.

Code:
10.2            Integer      Floats     Altivec
TiBook 1 GHz      100         100         100
G5 1.6            120         223         161

10.3            Integer      Floats     Altivec
TiBook 1 GHz       95          99          98
G5 1.6            103         224         159

I have tried running the test again to see if the numbers change and the integer does go up to 108 but that is it. The integer performance of a 1.6 Ghz G5 is equal to a 1 Ghz G4? Ouch. So hanibal was right.

[Eug]

Originally posted by Anand:
I have tried running the test again to see if the numbers change and the integer does go up to 108 but that is it. The integer performance of a 1.6 Ghz G5 is equal to a 1 Ghz G4? Ouch. So hanibal was right.

Like I said, I don't think we can really compare integer performance BETWEEN CPU architectures based on this single simple benchmark.

I just want to know what Skidmarks is doing so I can better understand why Panther integer is slower here than Jag integer.