People are placing far too much emphasis (at least on the home console side) on whether it's using a 14nm/Polaris chip. Nintendo's decision to use a 28nm/GCN 1.2 APU or a 14nm/Polaris APU (a decision which was likely made in late 2014, by the way), was based on the answer to the following question:
"For our target cost per chip, and out target thermal envelope, which will give us better performance?"
A 14nm chip will be able to clock higher within a given thermal envelope, and will likely get a bit of a performance boost via architectural improvements, but will likely still be more expensive per unit of performance than 28nm for some time yet. Hence why I still believe Nintendo will use a 28nm chip for the console, not because "it's Nintendo", but because it would actually provide
better performance than a similarly priced 14nm chip for Nintendo's launch timescale. That said, I'm not in a position to know for sure what the answer to that question is, and particularly not what AMD and Nintendo thought the answer might be when they made the decision in late 2014.
Pretty much hit the nail on the head. I know that you still highly doubt 14nm though, so you're probably a bit less confident than I am.
While I think a 14nm chip for the home console is quite unlikely, I actually wouldn't completely rule it out for the handheld. There are a few reasons for my line of thinking on this:
- Dropping from 28nm to 14nm will give a much bigger boost to performance in a very tight thermal envelope (i.e. a handheld) than a high thermal envelope (i.e. a home console).
- On low-yield fabrication nodes, die failures increase exponentially with die size, which means small dies tend to become feasible before larger ones do.
- With Nintendo seemingly moving towards cross-developing games across home console and handheld, this job becomes easier the closer the two devices are in performance (i.e. the higher the performance of the handheld). Hence there's an incentive in reduced development costs to increasing the performance of the handheld this time around (usually the incentive is the opposite).
- Similarly, Nintendo would have a much easier time cross-developing games for both their handheld and home console if they use the same GPU architecture for each.
- With AMD very likely to provide the APU for the home console, this would mean using an AMD chip for the handheld.
- It's unlikely AMD would be able to put out a competitive chip at such a low TDP on 28nm.
- Conversely, AMD should be able to provide reasonable performance within such a TPD with a 14nm chip.
I wouldn't bet any money on the handheld using a 14nm SoC, but it's somewhere within the realm of possibility.
That's incredible information you two! I wouldn't have thought it'd be that close or exceeding them already. Any newer information/expectations that Nintendo will go this route instead of x86?
There's no reliable information one way or another, but many of us are assuming that they'll go with ARM for the console on the basis that (a) there are viable cores available for them like the A72 and (b) it would allow them to use the same ISA for home console and handheld, simplifying tools and OS development.
Well, then you also have to throw away any discussion on NX even being close Xbone, since Nintendo making a 80+W console after Wii (15-20W) and Wii U (30-35W) is laughable.
There are two flaws with this line of thinking.
The first is that Nintendo would never release a console with high power consumption. The truth is that after the failure of the Wii U we have no idea what route Nintendo will take, and Kimishima has explicitly said that NX is not a follow up to Wii and Wii U, so it's entirely within the realm of possibility that they've changed course and are now willing to target a 100W+ power draw for their home console to compete with Sony and MS.
The second is that using a 28nm chip means that a console with XBO performance levels has to have XBO power draw. Firstly, modern 28nm processes consume much less power than the early 28nm processes used by PS4 and XBO. It's not quite the same as a jump to a new node, but there are surprisingly large savings to be had. Secondly, XBO used a narrow, high-clocked architecture which is inherently going to provide worse performance per Watt than a wide, lower clocked architecture. If Nintendo used a, let's say, 16 CU GCN 1.2 GPU clocked at around 600MHz, dropped the eSRAM and DDR3 in favour of a single pool of 256-bit LPDDR4, used A72 cores instead of Jaguar and dropped the optical drive and hard drive, they could put out a console with similar performance to XBO at a vastly lower power draw without going anywhere near 14nm.
