Fascinating results of some serious testing from a reputable source. We've had hints of this in the past. The G5 is fundamentally not that much faster than the G4, and real-world performance is not processor bound. Performance is affected by much more than the CPU. There were significant design compromises to fit a G5 into the iMac, and they have performance implications. Disabling the iMac's power management features helps performance, but it may stress heat management -- and those fans will run. (Of course in Minnesota we can simply put the iMac in the attic -- no heating problems in winter there!).
BTW, I think a similar analysis of various Intel systems would show similar results. Performance nowadays is often about heat, system throughput, memory, hard drives, etc. The CPU isn't the big factor for a lot of functionality.
Implications? The G4 iBook and G4 eMac are very interesting alternatives to a G5 iMac. The major unknown is Tiger. If one wants to run Tiger, should one opt for a G5 iMac? My general rule is that when an OS upgrade is very important (like Tiger -- it's something I want), the best strategy is buy hardware that ships after the OS goes GA and that ships with the OS.
Since I want a new Mac now and I want to run Tiger, I may choose to shop around for a used G4 system, or a new eMac, and plan to buy a new system post-Tiger. Nobody should upgrade from a fairly recent G4 machine to an iMac or even a G5 tower.
... Using QuickTime Pro 6.5.1 and QuickTime Player, we export a high-quality 50-second DV file to MPEG-4 format. Source and target files are on the hard drive. We use the standard "Default" settings.
This is a good real-world test of system performance, and the results are surprising: Right out of the box, the eMac G4/1.25GHz outperforms the iMac G5/1.8GHz system at Apple's standard settings, and the lowly iBook G4 is right on its heels.
If you change Apple's standard Energy Saver options to get "Highest" processor performance, the iMac G5 will outperform the eMac, but there must be some reason that's not the default, and clock speed alone should give the iMac a big advantage.
Apple is using the G5's special "slewing" feature to reduce heat and power drain, and the result is a real bottleneck.
Conclusions
The iMac G5 is a wonderful system, and we'd rather pay a few hundred dollars over the cost of an eMac to get one, but all the Apple hype about the G5 falls a little short when you see the low-cost eMac, with its slower G4 processor, pushing the iMac G5 in performance. The eMac is actually faster in several real-world situations, and that raises some serious technical questions that long to be answered.
In the meantime, you'll be getting a high-performance bargain with either the eMac G4 or iBook G4, and you still can't go wrong with the iMac G5.
If you want the ultimate in performance - or maybe just a super-large screen - the Power Mac G5 is the way to go, although we have some concerns about reliability with the liquid-cooled 2.5GHz model and would probably stick with 1.8- or 2.0GHz systems.
PowerBooks are nice, but pricy. The biggest advantage you get for the extra cost of the 12" PowerBook is the ability to drive a larger external screen in dual-display mode (up to 2048x1536), although the built-in screen has the same 768x1024 resolution as the iBook. [jf: the iBook hardware supports driving an external desktop, but Apple disables this -- possibly for heat reasons, possibly to protect PowerBook sales.]
The 15" PowerBook is an ideal mobile machine, and it can drive a big external screen on a desktop or use FireWire 800 to get disk performance more on par with a desktop computer's. This laptop costs almost twice as much as an iBook G4, however, making it an expensive option for part-time portability, and it's not as compact as the jewel-like 12" models.
The 17" PowerBook strikes us as an expensive alternative to the iMac with better portability and battery power.
One last factor is the G5's support for 64-bit processing, which is supposed to get a boost with next year's Mac OS X 10.4 "Tiger". Theoretically, this may be an advantage for G5 models, but the real-world advantage for general applications is questionable at this point.