I see the real message here being choice, the choice to have a big.LITTLE system, a 4PLUS1 system, an aSMP system, or even a good old fashioned basic DVFS/Sleep system.
It wasn't that long ago postings were whether more than 1 CPU could be used... today I have seen 8 cores being used in andriod, agreed these are not all 'big' tasks so whether you provide them a LITTLE cpu, or a cpu running more slowly with aSMP - the software complexity is pretty much the same.
Lets all just be thankfull we don't all need to eat a single DVFS based solution!
The NIH syndrome not withstanding, there is validity to Mr. Renduchintala's point about core optimization, that too application-specific core optimization where Q certainly knows its turf better than ARM.
@Rick: I too suspect there's debate inside Qualcomm about licensing its cores, it is not just the revenue but proliferation of its technology.
I'd not dismiss the big.LITTLE approach, though the challenges in supporting this scheme in software and OS level are far from trivial as pointed out. Using the ever-increasing (but hard-to-power) number of transistors to add more specialized processing cores is also a great approach and easier to support in software/developer level.
There's a great paper called "Is dark silicon useful?: harnessing the four horsemen of the coming dark silicon apocalypse" by Michael B Taylor of UCSD, where two of the "horsemen" (the "dim horseman" and the "specialized horseman") correspond to two of the approaches here. Recommended reading for getting a better feel of what may lie ahead for the processor world.
Blog Doing Math in FPGAs Tom Burke 13 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...