The chart stinks:
Photoshop, 150Mb file: G5@2GHz=51sec, G5@1.8GHz=76sec?
What, truth hurts?
That said, a fair bit of blame can rest on OS X here. Not for "processor optimization," but basic architecture. Every time Photoshop redraws its window while/after a filter completes, for instance, Quartz steps in and bites off another chunk of CPU time, Extreme or not. After 10 redraws (and a couple associated dialogs, one assumes), that starts to add up. I'm pretty sure the weird kludge of the kernel itself adds some overheads, and so on. (Windows, giant bag of filth or not, has had time to "evolve" various tricks around certain performance penalties; the system's been pretty well profiled by now, if only by sheer popularity and use.)
Does anyone know how Darwin works? Can anyone keep track?Throw out the Premiere results if you want, boggle at the Quicktime ones - as what should be a pure computational benchmark, the gap should narrow there; in that case, I expect there's some Altivec mis-optimization along with the OS overheads - ignore Word, and you're left with a picture of IBM
beginning to close the performance gap with what is, frankly, the best-performing desktop chip on the planet.
Now, Mac users don't want to throw out Quartz, at least, so some of this is a 'real-world' penalty
they have to live with. But clock for clock, the 970 is pretty close (look how bad the P4 shows in comparison, at 3.4GHz!), and if IBM gets to scaling it, it should certainly play well against everything else on the market. K8 is relatively mindblowing, P4 is a dog, and 970 is in the middle, while closer to the AMD side of the spectrum for IPC. (Basically, you can look at the G5 as an attempt to "P4-ify" PowerPC for 'big numbers' and ease of scaling... but in so doing, IBM couldn't stuff up nearly as bad as Intel.)
Again: K8 rules. It has a lot of circuitry dedicated to doing the 'right thing' even in spite of the x86 instruction set, and as the numbers show, it does that right thing really honking well. 970 dedicates a fair bit of circuitry to doing the right thing in spite of some assumptions made with PPC, and it does about as good, without having required quite the "full-on assault" it took to produce Hammer, or even - if I remember the pictures correctly - completely packing its die.
