Back in November, Anand talk to ARM about 64-bit. Arm said the A57 has more performance than the A15 but also uses more energy. Without 20nm, the A57 would have to be clocked much slower than the A15. Anand believed that would limit the A57 designs. Qualcomm today with the announcement of the A53 octa-cores proved Anand was right. The A57s are too hot to use at 28nm. It also shows that there will be very limited production of 20nm this fall and Qualcomm will more than likely not have a 20nm AP till 2015. At todays press conference, Qualcomm said that the A53 designs would be a S600 replacement as most applications do not use 8 cores. This leaves Intel's quad-core Moorefield as the only true top tier 64-bit Android SoC for the fall. This fall will be very interesting.
The reason that al the current ARM parts are A53s is because many of the vendors are working on their own custom cores for the high-end, such as the custom core used in the Apple's A7 and the Denver core used in NVIDIA's Tegra K1, both of which are 64-bit cores. Apple was the first to production and NVIDIA will be in production this fall as well. Note that some vendors are also working on big.LITTLE configurations with the A53 and A57. You will also likely see more announcments in the coming months. And you are correct that Intel's Merrifield is also 64-bit, but not the only one. The real question is when will we see a 64-bit version of Android from Google?
Regarding the type and number of cores, there are many ways to build an SoC and many applications beyond just smartphones and tablets. Is 8-core overkill? At the time it may be, but then again, so is 64-bit without the applications. In fact, it is unlikely that you will be abl;e to use more than 4MB of memory in most handsets due to the size and power limitations.
jpfj - I think we need to log that one with the many other memorable industry quotes that were rather short sighted. Obviously, they would not spend the time to spin the silicon if their was not a customer request. Everyone is looking for some form of differentiation and that is leading to a plethora of SoC solutions.
JimMcGregor, an interesting point you raise. Rather tangential, but do you think customers request features from silicon vendors and these are then architected by the vendor, or do you think silicon vendors try to read the market and architect things that they believe their customers want and then try to hard sell it to them?
My experience is more with the latter, but am no way saying that this is the way the industry works. I'm curious, what's your view?
tpfj - Customers, referring to both handset OEMs and carriers, do have specific requirements of their silicon providers. One request that has arisen from Asia is the increase in the number of cores. OEMs are still searching for a way to communicate performance and newness, and they like simple numbers like frequency and the number of cores. And who can blame them, the industry has programmed consumers to look for these numbers. In addition, the number "4" is deemed unlucky in Chinese custom, while the number "8" is deemed lucky, which is another factor OEMs don't face in other regions. As a result, look for a wide variety of solutions in the market that are aimed at not only cost and performance, but also marketing factors.
I don't think anyone has tested A57 yet so that conclusion is not substantiated by evidence. AMD will have 8-core 2+GHz A57 SoCs on 28nm soon, and despite being a server part it uses less than 3W per core. A mobile version would use far less power without needing to be clocked at a lower frequency.
Anand seems to dislike A15, presumably because it beats Silvermont by a good amount. Given it will be running at 2.1-2.3GHz in mobile phones (at 28nm of course) the claims about A15 not being power efficient are false!
Also note NVidia K1 will be out at the end of the year, so even if A57 doesn't make it into mobiles by then, there will be high-end 64-bit ARM mobile SoCs available by then. It's quite obvious that Moorefield will not stand a chance - A57 is about twice as fast, and Denver will be faster still... This fall will be very interesting indeed!