Sorry
@Thraktor , I'm probably being dense (it's the end of the week), but are you arguing that TSMC 4N makes more sense, or that it's not worth shrinking from Samsung 8nm?
No worries. What I'm saying is that we can infer what manufacturing process they are planning to use based on the fact that they chose such a large GPU.
If they were planning to manufacture on 8nm, then I don't think there's any reasonable justification for them choosing a 12 SM GPU. They could achieve
better performance in portable mode at lower cost by using a smaller GPU (eg around 6 SMs) and clocking it higher. Therefore I think it Samsung 8nm is very unlikely. Conversely, 12 SMs would seem to be a very sensible design for TSMC 4N, so I think 4N is more likely. I should also mention that I don't think there's any reason to believe that they've changed manufacturing process at any point. They wouldn't have designed a 12 SM GPU for 8nm and only after the fact realised that it's far too big for Nintendo's requirements. I could have told them that, and Nvidia know far, far more about this kind of stuff than I do.
There is one additional caveat to this, which is that I'm assuming that they'll be using all SMs in portable mode. It's worth noting that it's technically possible for them run fewer SMs in portable mode than docked. If we are wondering
why they'd do such a thing, the only reason I can think of is if they want to have a much larger performance gap between portable and docked mode than on the original switch (eg 4x or more). This isn't really plausible if you use the same number of SMs in both modes, but if you effectively add extra SMs for use in docked mode, then you get a larger than expected GPU that has a much bigger gap between docked and portable performance.
This isn't something I expect, though. Although we will see a bigger gap in output resolutions than before (I'm expecting 4x between a 1080p handheld display and 4K TV out, compared to 2.25x between the 720p screen and 1080p TV on the original Switch), I don't expect rendering resolutions to actually be 4x higher in docked mode. Part of this is because the hardware seems to be designed around DLSS, and DLSS tends to perform better with higher input resolutions. That is, although you might be running DLSS with a 1080p internal resolution and 4K output in docked mode, it's not necessarily the case you'd want to run at 540p internal resolution scaled to 1080p for portable mode. Something closer to 720p to 1080p scaling would probably give closer image quality (in terms of sharpness relative to resolution, and lack of artefacts) to the 1080p to 4K scaling gets you in docked mode. Another reason ties into
@oldpuck's post on the previous page, which is that I'd expect more traditional CPU tasks to run on the GPU this time around due to the existence of the tensor cores. These tasks don't scale with resolution, so depending on the ratio of GPU time spent between compute tasks to rendering tasks would actually want the gap between GPU performance across the two modes to be smaller than the gap in resolution.
Going with an extra-large GPU and disabling half of it in portable mode would also mean Nintendo would be paying extra for hardware they're only using half the time, and there's extra software work required both to implement the switching between modes in a clean manner, and also on the side of game developers to account for the different available hardware in each mode.