I wonder if Nintendo would go with an APU which would likely cost significantly more than the rather outdated silicon that they've used in their handhelds so far. Especially after the price issues they had with the 3DS. At least I assume that it will be more expensive given the die size and the node process. Maybe they could go for a small derivate with only two Puma cores and a smaller GPU.
They'd swap the Puma cores out for ARM, probably A53's (which take up about 1/3rd the space of Puma). And the 3DS price issues were much more to do with the 3D screen and Nintendo's desire for day-one profitability than silicon costs. Besides, 28nm is a very mature node now, and a ~100mm² 28nm die wouldn't be completely out of the question for a $199 break-even handheld (unless they do anything crazy like throwing a donut-shaped screen in there, which isn't at all out of the question).
You do have a point. That does not mean AMD are not deeply sorry about that : )
Well, I'm sure AMD believe they would have managed it fantastically had they held onto it, but I'm sure they also believe that K12+Polaris will see them gradually dominate the Android world, so I'm guessing the two delusions would cancel each other out.
That's a tablet SoC, not a hh SoC - 4.5W is still way too much for a hh, let alone one by nintendo (who never go reckless with their TDP). And the problem is, cutting the CPU cores won't be enough - they'd need to downgrade the GPU in there.
Perhaps a little, but I don't think they'd have to gimp the GPU all that much. For one thing, Mullins was early-2014 28nm, whereas an NX handheld would be late-2016 28nm, so you've got two and a half years of process improvements to rely on to improve efficiency, which can count for a lot. Secondly, Mullins was using GCN 1.1, whereas a modern chip would presumably use 1.2, which is considerably more power efficient. See Tonga's significant perf/W improvement vs. Tahiti on an almost identically sized die, or the power efficiency of the Fury line (though HBM plays a part in that one). A sort of GCN 1.2.5 incorporating aspects of Polaris wouldn't be completely impossible, either, given that the uarch R&D for Polaris would have been well underway when design started.
I'm not saying that a modern Mullins-alike would necessarily be any kind of powerhouse in a handheld TDP, but it would satisfy what I feel would be two important criteria by Nintendo's standards:
1. It would be a significant jump over the 3DS
2. It would provide predictable scaling for cross-development with a GCN-based NX home console
Given the way Nintendo's been talking about the NX, I don't think an AMD 28nm APU for the handheld can be ruled out.
