Well, mine wasn't either ; )
QCOM is a turn-key phone chipset provider - and that is a major advantage of their business - here we completely agree. But that's not their sole advantage in the industry. QCOM are also one of the few ARM architecture licensees who can (and traditionally do) make better uarchs than ARM themselves. And one of the top-four mobile GPU IP holders (Rogue/Tegra/Mali/Adreno). They make a myriad of Adreno + A53/A57/Kryo SoCs as we speak. On all relevant fabnodes - apparently the mobile GPU landscape might change come Polaris and Samsung GF's 14LPP. But QCOM would be among the top candidates for a semi-custom hh SoC design from the POV of a platform holder, nintendo or otherwise.
Their custom ARM cores are certainly competitive, but the only announced Kryo-based chip is the 14nm 820, and it doesn't seem likely that they'll be providing it on a 28nm process. Krait is an option, and while it was a good choice against ARM's 32-bit designs, Qualcomm have dropped it in favour of stock ARMv8 cores in their 28nm 600 series, so I'm not sure there's any reason it would be used here.
In the event that AMD are providing only the home console SoC, then I wouldn't rule out Qualcomm, but I wouldn't see any reason to put them top of the list. In fact, the one option that we're missing out on is that Nintendo licences a suitable ARM core, a GPU design, and whatever other IP they may need, and simply designs the chip in-house. It's not out of the realm of possibility, as the amount of custom IP they'll need is minimal, and it seems to be pretty much what they did with the 3DS (although in that case the CPU and GPU may have been on separate dies?)
Ok, so you're arguing from a GPU ISA-centric POV, I understand. I might have agreed with you if it wasn't for the concessions that nintendo would need to do to achieve that.
It's not just the ISA, although that's where we've got the most readily-available info on the extent of divergence of the architectures. My belief more generally is that, from a developer's point of view, the statement "This technique is efficient on the home console, therefore it should be efficient on the handheld" isn't going to be any more true for Adreno with a GCN home console than it is for, say, Rogue or Mali. And it's the ability for developers to make that statement that I feel is the biggest argument in favour of using the same architecture between console and handheld, even aside from any benefits to Nintendo in simplifying tools development, support, etc.
Well, let's wait and see : )
This is our main issue here. If history has taught us anything, it's that if there's one thing you can predict about Nintendo's hardware design, it's that it's not going to be what you predict!
Regardless of devs willingness to write asm most systems are already moving away from giving direct low level access, instead wrapping it with an intermediate level api. So "much more predictable toolchain behaviour" as you say is exactly why I'd still go for it, predictable performance scaling across multiple performance levels in an otherwise unified platform imo is a must. While an abstracted api layer can ensure compatibility with different uarches, it won't be able to make performance predictable.
Nintendo already showed early eagerness to make use of GPGPU even having picked a for this purpose outdated VLIW based uarch for Wii U, with NERD reportedly working GPGPU related technologies. Iwata mentioned NX would build upon Wii U which I'd expect to refer specifically to the GPU capabilities. Keeping PowerPC and using that on handheld is very unlikely. AMD being involved for offering an x86 based CPU for both form factors imo is even more unlikely, and AMD keeps referring to its plans to eventually offering ARM based solutions as well. So that with the news that AMD got a major design win makes me expect NX' GPU being another result of the long standing partnership with ArtX/ATi/AMD.
Right. My current guess is that NX will be first released as a GCN+ARM based console (potentially additionally including Wii U's 3 core PowerPC for backward compatibility with some platform software bridge so much of the eShop software works on both) and later using the 14nm Polaris shrink for handheld NX and more powerful console revisions which would all be compatible and release in an about yearly fashion similar to Apple practices. If the handheld NX is planned to be released together with the home console NX I don't expect them to be on sale before 2017.
I wouldn't rule this out, except that I wouldn't expect such frequent refreshes (every two years at tops). The only issue with releasing the home console in 2016 and the handheld in 2017 is business more than technology. The 3DS is significantly older than the Wii U, and although the Wii U isn't selling great, it is actually marginally increasing year on year. The 3DS, by comparison, dropped pretty hard last year, and would be barely scraping along in 2017 if they don't replace it until then. With handhelds historically accounting for much more of Nintendo's income, if either device is to land before the other you'd have to expect them to push the handheld out the door first. That means the handheld this year, which means it would most likely have to be 28nm (unless they're willing to take the hit on early 14nm).
We are only talking 28nm nodes or 14nm nodes. Isn't there anything in between that Nintendo could choose from?
There's 20nm, but it's getting very little use outside high-end mobile SoCs, and AMD (who we pretty much know are making an APU for at least one of the NX form-factors) aren't using it at all, jumping straight from 28nm to 14nm.
