I have been bearish on ARM's (ARMH) Cortex A15, believing that it is too power hungry for the mainstream/performance smartphone lines that are so popular. While custom designs such as Qualcomm's (QCOM) Krait have been able to more adequately service this market, many of ARM's partners do not have the resources/inclination to develop their own processor cores, and as a result are left out in the cold. ARM recently announced its Cortex A12, which is supposedly much faster than the prior Cortex A9 but consumes much less power than ARM's own Cortex A15, which will enable partners to develop competitive mid-range smartphone system-on-chip designs.
Cortex A12: Perfect For Marvell, Nvidia, And Others
The Cortex A12 moves to a dual instruction issue pipeline rather than the very wide triple instruction issue pipeline found in the Cortex A15. While I suspect that ARM's intentions with the A15 was never to really power smartphones (or to do so via big.LITTLE), both big.LITTLE and A15 have proven problematic for these segments, as Samsung's (GM:SSNLF) Galaxy S IV (international edition) has shown. The edition powered by Qualcomm's Snapdragon 600 has superior battery life and good performance characteristics, proving that maybe big.LITTLE may have been a better idea in theory than in practice.
Anyway, so in a modern micro-processor, while a wider issue improves the number of theoretical instructions per cycle that a processor can execute, in practice, it is very difficult for a modern processor to come even close to hitting this peak theoretically throughput...so it is usually not particularly power efficient to have a very wide front end, especially as this necessitates a very wide back-end, further threatening the delicate performance/watt balance. This is why both Intel (INTC) and AMD (AMD) have focused on dual issue designs for their "Silvermont" and "Jaguar" cores, respectively, and have still managed to hit excellent performance numbers despite narrower front ends.
So, the Cortex A12, which unfortunately will not be with us in devices until late 2014/early 2015, solves ARM's problem of having nothing in the mid-range of its product stack for phones. While Intel has chosen to optimize its low power cores to have a very wide dynamic range, ARM has chosen to focus on hitting different power/performance targets across various processor cores. Both approaches seem valid, particularly as it is likely that from ARM's perspective, it is better to have several designs that are as small and as efficient from a raw transistor/manpower to design standpoint, rather than putting all of the effort into one core that is hard to segment, and probably requires much more work to validate.
Further, ARM's licensing/royalty model likely calls for lower royalties on the much smaller "little" chips, much higher royalties on the "big" chips, and then somewhere in the middle with the mid-range...erm..."middle" chips. This is a win/win for ARM and its customers.
I expect to see this A12 core in mainstream designs from Marvell (MRVL) and Nvidia (NVDA) (the latter will be using its own custom core in its higher end/higher power designs), as well as some of the other more cost-conscious Asian SoC vendors such as Allwinner and MediaTek.
I think big.LITTLE is proving to be a bit too problematic for the majority of performance (but not uber-high end) smartphones, and believe that ARM has done the right thing by aiming for a good compromise between the power hungry, higher performance A15 and the lower performance, lower power Cortex A9. This should enable the weaker ARM partners, and those whose expertise is not in CPU design, to stay competitive in an increasingly competitive market.
Additional disclosure: I may go long MRVL at any time. I am short ARMH