Apple's A7 is largely seen as a very impressive piece of CPU engineering. However, what you don't see is how much software makes a difference; apple's A7 is impressive (and A6) mainly because they leverage Apple's substantial software advantage rather than great hardware prowess.
It can be seen that a lot of improvements in performance across generations are from software upgrades rather than any sort of hardware improvement. Case in point, the improvement from the iphone 5 to the 5c.
AT release article of the iphone 5
http://www.anandtech.com/show/6330/the-iphone-5-review/10
The difference between these benchmarks is that one was taken with IOS 6 and the other with IOS 7. As you can see, the iphone 5 improved by 40%.
Likewise with Octane
71% improvement from IOS 6 to IOS 7.
Sunspider shows a smaller but still present improvement.
I have no doubt that the user experience from the A7 is every good as the charts indicate but the actual hardware performance isn't nearly as close to the baytrail or S800 as benchmarks indicate. Apple's software optimizations allow them to put a very strong showing with weaker hardware. (Geekbench is not used in this comparison as it focuses too much on encryption and tends to operate on single buffers of data). As shown in other benches, 5s performance is notably higher than 5/5c performance; my point here is that the actual hardware is less powerful in comparison to other hardware than the benchmarks indicate.
This is the problem with using JS benchmarks.
It can be seen that a lot of improvements in performance across generations are from software upgrades rather than any sort of hardware improvement. Case in point, the improvement from the iphone 5 to the 5c.
AT release article of the iphone 5
http://www.anandtech.com/show/6330/the-iphone-5-review/10


The difference between these benchmarks is that one was taken with IOS 6 and the other with IOS 7. As you can see, the iphone 5 improved by 40%.
Likewise with Octane


71% improvement from IOS 6 to IOS 7.
Sunspider shows a smaller but still present improvement.
I have no doubt that the user experience from the A7 is every good as the charts indicate but the actual hardware performance isn't nearly as close to the baytrail or S800 as benchmarks indicate. Apple's software optimizations allow them to put a very strong showing with weaker hardware. (Geekbench is not used in this comparison as it focuses too much on encryption and tends to operate on single buffers of data). As shown in other benches, 5s performance is notably higher than 5/5c performance; my point here is that the actual hardware is less powerful in comparison to other hardware than the benchmarks indicate.
This is the problem with using JS benchmarks.