I'd like to refer to some of Anand's observations with the occasional quibble (though nothing serious - this is a solid piece of work, which I admire a great deal):
1. Anand's tests used a default memory configuration of 512MB. He points out that memory use of the Intel Macs seems slightly higher (see page 5 of the report), and particularly so for Rosetta apps. I made this point in a previous entry, and am glad to see it confirmed. While it is indeed fair to compare the 2 machines with the same config, I would still guess that extra memory would benefit the Intel iMac disproportionately - especially for Rosetta-based tests. If you intend to buy an Intel iMac and do serious work, do not consider anything less than 1GB, preferably 2GB. It would be interesting to do the same comparisons with both machines more heavily RAM-loaded.
2. Partially as an aside, Anand shows what has oft been mentioned, but never actually resolved: that the iMac G5 default setting for Automatic for Power consumption significantly lowers the performance of many of the tasks performed (example provided was H.264 conversion which took over 25 minutes with Automatic on, and 12.3 minutes with Automatic off, even though theoretically the processor should have adjusted. The Intel test was done in 9.8 minutes incidentally). All of Anand's tests were ultimately performed with this setting OFF, allowing the G5 iMac to perform at its highest levels. iMac users who have not changed this setting would experience considerable extra performance improvements in moving to an Intel iMac (or they could turn Automatic Off!).
3. Perhaps the most interesting finding is on page 7 where Anand looks at performance per watt by measuring power usage for the whole system plugged into the wall. Remember this is the primary reason Steve gave for the switch. The Core Duo uses about 2/'3 the total power of the G5 iMac at both idle and full load. And, this is total system power - including other components - especially the screen. Given that these other components are by and large the same (Core Duo/G5), the contribution to the drop in power usage due to the chip is possibly 2x or better. (Incidentally, for a 24x7 operation, the saving in electricity is likely to be around £15-20 per year). Applying these figures to a particular performance test for a native application (a Quicktime H.264 encode) actually demonstrates a 90% performance per watt improvement over the G5, according to Anand's tests.
4. Is this (QT H.264) test fair to either machine? Anand says "You should also keep in mind that the Quicktime encode test is actually one of the Core Duo's weakest tests, so the actual performance per watt advantage may be even higher in tests where the Core Duo has a larger performance advantage over the G5. This is just a good test to show power consumption since it is fairly CPU bound and consistent in its CPU load."
5. On other real world tests (native applications) the Intel iMac is generally considerably faster. In tests where the applications are not multi-threaded, the performance is much closer - indicating one of the cores closely resembles the G5. But when both cores are used, there is usually a substantial difference. Remember these are real world tests.
6. Once again, I would note one weakness in these tests - there is no parallelism. While Anand identifies tests where there are multiple threads which would take advantage of the Core Duo, he does not run tests combining one application with another. I still think this would be a valid test to run. For example what's the effect on browsing on either machine when an H.264 export is being done? (Given the number of tests performed I'm not really complaining - just identifying another area that would be of interest).
7. The Rosetta results are a bit disappointing - especially compared with some others I've seen. Anand finds performance frequently at 30% of G5 level. However, I think he is stressing it quite a bit - very large Word documents for instance. While I question what he uses to determine memory footprint, he does point out that with a 116 page word document open, memory use on the Intel iMac is about 3x that on the G5. This to me indicates that many of his Rosetta tests are suffering from low memory. It would also be interesting to do these tests on a machine while it was doing an H.264 conversion. Nevertheless, Rosetta's invisibleness is a positive, and even these performance numbers are pretty good when considering that this is emulation.
8. Anand provides some performance numbers for the Core Duo with one processor turned off. Most of these show the chip a reasonable match for the G5 on many native applications. This bodes well for (likely) single core Mac minis and iBook replacements.
9. And, I still think the $64,000 question is how do these Macs compare with similar Windows machines at everyday tasks - iTunes encoding, H.264 encoding etc. Are we just waiting for real shipments of Windows-based Core Duo machines before these tests are available?
But all in all, I think this is yet more evidence that these machines are fast - and represent a leap in performance in a single machine iteration that we haven't seen for many years. Providing the apps are native, that is. If you are a fulltime PhotoShop user (in which case you'd probably not be using an iMac) or are regularly editing large Word documents, then perhaps you'll want to wait a while longer.
But let's leave some of the concluding comments to Anand:
I like the iMac, I like it a lot...It took me this long to look at it, but I think it could quite possibly be Apple's strongest offering as it accomplishes exactly what they are trying to do - which is build lifestyle computers. ...
...Just about every application I'd use is already available as a Universal binary, the only exceptions being anything from Microsoft or Adobe/Macromedia...
...It's also worth noting that a pair of G5s could never make it into an iMac of this form factor, meaning that the Core Duo's dual core performance advantages are reasonable to flaunt...
...so I tend to believe that things will get faster for Intel based Macs over time. Not only when Rosetta is no longer needed, but also as applications are better optimized for their architecture (e.g. Quicktime)...
...I'll close, as always, on a note about the future. We've seen that today, Intel already has the performance per watt crown with the Core Duo, and they also have the power advantage, consuming a third less power than a similarly clocked G5. Yet the first Intel based Macs are nothing more than the G5 versions with a different motherboard and cooling. You tend to not over design your chassis when you are Apple, you design them to be as sleek and as minimal as possible. With the Core Duo based iMac consistently consuming 20 - 30W less than the G5 version, you can expect that the truly exciting Intel based Macs are the ones that don't look like these. It's those that I would personally wait for.
Incidentally, Anand makes no mention here that Low Voltage and Ultra Low Voltage Core Duo chips will also become available over the next 2 quarters giving Apple even more flexibility in product design while maintaining G5 beating levels of performance (1.83GHz in the Low Voltage; down to 1.06GHz in the Ultra Low Voltage design). So, the first fruits of the Apple Intel partnership look very sound and 2 year old G4 users would likely experience a significant performance leap. But the really exciting products may be for round 2.
Great work AnandTech!