If AMD get into SoC now, I'm guessing they are still a few years away from having the parts ready for retail. None of the ARM licencees have had particularly fast ramps, so it's probably unrealistic to think that AMD are going to jump in and produce a workable SoC in the immediate timeframe.
I don't think anyone is expecting AMD to drop desktop/server CPU in any immediate future -at least not until they have APU and the SoC's earning revenue, and that is going to take time. My thinking is that if AMD's R&D is financially constrained (likely, no?) then how do you divide up research? If AMD are moving to SoC as seems lightly they need to apportion R&D funding...where is the cash coming from? AMD can't strip the graphics division (since it is both profit and is tied to APU), that pretty much means desktop/server starts losing development funds* -so I would think it a logical leap to see AMD mirror K8/K10 progression with Bulldozer/Piledriver/Steamroller etc. i.e. squeeze/refine the existing design as much as possible without requiring a major redesign and influx of research funding. As far I'm aware, AMD's future announced CPU's are all based on the Bulldozer architecture, so it would seem that incremental gains are more the future for AMD than major overhauls and ground-up redesign.
AMD's own literature talks about a 40-50% performance improvement between BD and Excavator -
assuming 0% hyperbole- how likely is that to be competitive against what Intel are likely to be fielding extrapolating from what is here, and what is known re: performance (single/multi thread)**, power envelope, die size.
Intel have the luxury of cash and a vision of x86 on itty-bitty chips. Sandy Bridge (new arch)-> Ivy Bridge -> Haswell (new arch) -> Broadwell -> Skymont (new arch) ->Skylake...whether or not that x86 SoC do-able this way -or with Atom, who knows? what is known is that Intel can be fairly relentless in pushing tech (particularly anything originating from themselves) and they have the marketshare to shape the market to a fair degree. I think AMD realize that they can't match Intel design-for-design, especially when you take into account that Intel have such a large lead in perf/mm and perf/watt...and one huge advantage in that Intel's foundries and process are tailored to Intel's design. At the present time, AMD have to contend with Intel on a µarch-to-µarch, whilst simultaneously being reliant upon GloFo to match or beat Intel in foundry process while having to migrate to gate last and play catch-up with FinFET or any other esoteric transistor optimization...and of course the longer they stay on this road, the harder it will be if they fail and have to start from too far back against ARM.
Reasonable assumptions ? I'm open to hearing other theories, so hit me!
*Assuming profit lines, marketshare and ASP's/margins stay relatively stable
** If AMD knew that BD was going to almost entirely multithreaded dependant, why didn't/hasn't AMD's Gaming Evolved program pushed/funded development of games to use BD's strength? (i.e. not just the occasional MT game but full x64 .exe's and coding that fully utilizes BD's ISA's)
Remember the conversation we had when I thought AMD was on a time-line for Soc?
(of course you do...you probably remember the thread number:haha
This one ?