The one thing this doesn't account for is that those scores came from all sorts of different Geekbench versions. In most cases they're probably much older than the current Apple ones. That means one could be compiled with compilers that are as much as 8 years older.
Geekbench is really not a good benchmark, it's so poorly regulated and reported :/
In retrospect, I feel the largest contribution by far of SPEC is how it helped bring awareness of compilers into focus generally, and in benchmarking particularly. It's unfortunate that its influence has waned.
In this case, using geekbench is better than the alternatives (Java and browser benchmarks). Otherwise we wouldn't be able to do any back of the envelope estimations
at all, and that would be pretty boring.
So - I went back and looked at Intels latest and greatest in low power processors. The 17W Ivy Bridge i5-3317U made on Intels 22nm Tri-Gate process. Making comparisons gets a bit rockier, due to Turbo technology among other things. I focussed on single threaded performance to ensure that the frequency would stay high (2.6 GHz), and also because comparing cores was what I was interested in for this exercise.
First things first: the overall Geekbench score is 5147, or just under three times that of Apples iPad4. So that is overall where it stands today using this benchmark, tablet vs. expensive ultraportable, and very roughly an order of magnitude difference in power draw.
If we look at the six single threaded integer benchmarks the average factor the Intel processor is faster is 2.37, or 1.28 adjusted for frequency
For the six single threaded floating point benchmarks the average factor the Intel processor is faster is 2.61, or 1.40 adjusted for frequency.
The memory subtests show a wider spread, but show a somewhat greater advantage for the Ivy Bridge, as expected.
The nice thing about these numbers is that they are consistent with the Core2Duo results, so we have no reason to be more suspicious than normal. And the indication is that Ivy Bridge has an architectural IPC advantage over Apple Swift, but that it is remarkably modest given the increase in complexity, roughly a third higher. What that implies is that Intel can't bank on architectural refinement carrying their day in coming CPU battles. (Indeed I'd be inclined to say that the experiment indicate that ARM has the architectural leg up, but that's a rather involved discussion.) Also, Intels advantage in process technology is factored in here.
Of course from a business point of view, the Intel Core i5-3317U has a tray price of $225, while the cost of the A6x should be roughly $20. I think there are some conclusions to be drawn from that and the power figures for the future, but I'll refrain for now, and allow tangey to get his in first.
