Here's a line for improvement: cut one CPU cluster from the hh and use a better GPU instead.
I think this is actually bad advice.
Alright, since everyone is a bit more reasonable with 14nm now, let me add to the discussion, first the Console will have much more die area reserved for the GPU than the CPU, but the die area on the handheld should be smaller, given that GPUs scale well with resolution but CPUs do not. Meaning if the console is 1080p and the handheld is 540p, the handheld's GPU can shrink by 4 times or even more, fairly easily. The CPU is performance locked somewhat, obviously it doesn't need to keep a 1:1 ratio of performance with the console, but it does need to be relative to some performance metric. I'm getting at the CPU in the handheld taking up a proportionally larger die area than the one in the console is taking in relation to their respective GPUs.
I see A53 thrown around a lot, it is a great cpu and I expect it, but I think the power consumption can be a bit misleading when compared to performance. 4 A53 cores clocked at 1.5ghz is the same power consumption as 2 A72 cores clocked at the same 1.5ghz, 0.8watts. My suggestion is that the handheld uses 2 A72 cores while the console uses 4, they both have at least 2 A53 cores, which at 1.5ghz runs 0.5 watts.
So give the CPU 1.3watts of consumption for the handheld would be 2 A72 cores and 2 A53 cores clocked at 1.5ghz, this I feel is better than 6 A53 cores clocked at the same, consuming the same amount of power. Developers don't want to use more cores, they want to use faster ones. In order for them to achieve a reasonable power envelope in the handheld, rather than 4CUs from polaris, they would be better served with 2CUs and more CPU power, which is basically my entire purpose in this post, the closer the CPUs are in the console and handheld, the easier development will be, we don't know polaris' power draw yet, but if they can fit 2 A72 cores and 4 A53 cores all clocked at 1.5ghz in the handheld and 2CUs, under 2.5watts for just the SoC. It would be the smart way to go and they can pump the console with 4 A72 cores @ 2.3ghz and 4 A53 cores @ 1.5ghz, performing very close to PS4's 8 jaguar cores, in a lot of cases out performing it.
My logic is basically that Nintendo should want to satisfy these goals:
1. Release a home console that is within touching distance of Xbox One and PS4, at a reasonable price
2. Release a handheld which is relatively cheap (i.e. $200 or under)
and
3. Make developing one game for both the home console and the handheld as simple as humanly possible
The first goal is relatively easy, but satisfying both it and the third goal is easiest if you low-ball the XBO and PS4 a little, i.e. something a little less powerful than the XBO (which should also keep the price low).
The second goal on its own is trivial. They could probably make a handheld that outperforms the 3DS (w/o 3D or anything like that) for $80 if they really wanted to. When combined with the third goal, though, they basically want to have as low a screen res as they can get away with (i.e. 480p/540p) combined with the most powerful SoC they can for $199, and without anything else that would push up the BoM.
Aside from the above, the third goal is most effectively met if they use the same CPU arch (i.e. ARM) and the same GPU arch across both SoCs. AMD and Nvidia are pretty much the only companies that could do both, but it seems like they're going with AMD. Therefore, the most straightforward and sensible approach for Nintendo to take would be a home console with a 28nm AMD APU with 12 GCN 1.2 CUs running at around 700-750MHz, and a handheld with a 14nm AMD APU with 4 Polaris CUs running at whatever frequency can be squeezed out of them. Even if they had to take an initial loss on the handheld it could well be made up for by reduced development costs over the lifetime of the device.
That said, Nintendo have never been the types to take "the most straightforward and sensible approach", so I'm personally expecting some kind of bewildering but fascinating new feature that happens to consume 40% of the handheld's budget, squeezing a decent SoC out of the picture.
Glad we are thinking the same way now, but I'd like to put out my own stab at the specs.
1. AMD has talked so much about Nintendo, and even the handheld successor to 3DS, I think they will be behind both systems.
2. I expect polaris in the handheld for it's power saving. I don't see the handheld exceeding 4 watts.
3. I think the console and handheld will both use a custom APU created at 14nm as it saves power and I expect AMD to be the designer anyways, also it is more simple for a single team to produce a single SoC family than 2 separate ones.
4. They will probably reasonably target current gen performance with the console.
Given those assumptions, I think the specs should quickly fall into place:
The console: $199
GPU: 8CUs @ 800mhz - 1ghz (819GFLOPs - 1 TFLOPs) at 1TFLOPs, Polaris should perform very close to XB1, though we don't know how well Polaris out performs GCN, we do know that it does, and should strike close.
CPU: 4 A72 cores @ 2.3ghz and 4 A53 cores @ 1.5ghz (up to 2 reserved for OS)
RAM: 8GB (I'm not going to touch on tech here, it doesn't matter too much)
The handheld: $199
GPU: 2CUs @ 500mhz 128GFLOPs (6 to 8 times slower than the console, but at 1/4th the resolution and a smaller screen, this should handle the same games with minimal rework)
CPU: 2 A72 cores @ 1.5ghz and 4 A53 cores @ 1.5ghz (offering the same OS experience could actually be important to Nintendo and would give them the best environment to tackle a faster OS)
RAM: 2GB
Whatever they do, they should spend more on the CPU for the handheld, to get the system closer to the console, hopefully this gives everyone some insight into how you'd combine these environments a bit more.