Die shot of the new A6.
it is assumed to show 2 cores and 3 GPU blocks.
Interestingly enough, what may not be obvious at first glance is that the die size of A6 (iphone 5) SoC actually
increased relative to A5 (iphone 4s, ipad 2) SoC, normalized for differences in fabrication process. In fact, the normalized die size of A6 (iphone 5) SoC appears to be virtually identical to that of A5X (ipad 3) SoC!
When moving from A5 (iphone 4s, ipad 2) SoC to A5X (ipad 3) SoC, Apple increased die size area by ~ 34%, where nearly all of the SoC die size increase went towards the GPU. When moving from A5 (iphone 4s, ipad 2) SoC to A6 (iphone 5) SoC, Apple increased die size by ~ 36% (normalized for differences in fabrication process), where approximately half of the SoC die size increase went towards the GPU and half went towards the CPU.
So as I mentioned before, the use of a smaller fabrication process was of extreme importance in shaping the end result for iphone 5. Due to the dramatic increase in transistor density when moving from 45nm LP fabrication process to 32nm LP fabrication process, Apple was able to increase normalized GPU die size by ~ 50%, and was able to increase normalized CPU die size by ~ 50%, relative to iphone 4s. Due to the very large reduction in leakage current when moving from 45nm LP fabrication process to 32nm LP fabrication process, Apple was also able to increase GPU operating frequency by ~ 33%, and was able to increase CPU operating frequency by ~ 50%, without increasing power consumption relative to iphone 4s. So the combination of increasing GPU/CPU normalized die sizes by ~ 50%, and increasing GPU/CPU operating frequencies by ~ 33-50% goes a long way in explaining the performance differences between iphone 5 and iphone 4s.
When comparing a [1.5GHz] dual-core Krait SoC vs. a [1.2GHz] dual-core A6 SoC, the overall CPU performance score is actually pretty similar in Geekbench, within ~ 5% of each other. At these operating frequencies, the dual-core Krait SoC has ~ 10% advantage in floating point performance, while the dual-core A6 SoC has ~ 15% advantage in memory performance, and ~ 20% advantage in integer performance. Of course A6 does have a large advantage in memory bandwidth performance, but that is not weighted as high as floating point, integer, and memory performance in Geekbench.
When comparing a dual-core Krait SoC with Adreno 320 GPU vs. a dual-core A6 SoC with SGX543MP3 GPU, the overall GPU performance is also pretty similar in GLBenchmark 2.5.
It will be interesting to see how power consumption compares between dual-core Krait SoC's (with Jellybean and Windows 8 operating systems) vs. dual-core A6 SoC (with iOS operating system). As long as the power consumption of dual-core Krait SoC's with Adreno 320 GPU is reasonably competitive to the A6 SoC, then these two SoC's will more-or-less have CPU and GPU performance parity and performance per watt parity.
