>>> As it's kind of "mobile" version of G5, it was perhaps silly to
>>> expect it to perform like [...] G4 per Mhz...

>> Why? The G4 up to 1.67 GHz was used in mobile devices as well.

> Yes but G4 was designed for desktop.

...yet is at 1.67 GHz obviously low-power enough for mobile usage, so its characteristics can't be too far off a dedicated mobile chip. So why was it silly to expect the PA6T at 1.8 GHz, that is two years younger than the MPC7447A/B, to be able to beat the G4 in per-clock performance?

> it has shorter instruction pipelines

True, shorter than PA6T. But then, PA6T and G5 have the same length, yet G5 has better per-core performance than G4 in most cases (as long as AltiVec is not involved that is), so this isn't very convincing.

> and better Altivec

True, G4 AltiVec has best per-clock performance, but I'm referring to non-AltiVec performance (see for instance Dhrystone benchmark which doesn't use AltiVec at all).