Maybe UFS (internal ) and CFexpress card(removable).

1) Wouldn't the likleyhood be that backwards compatibility would be no-go if they switched GPU manufacturer anyway? Are there recent examples of BC between consoles that don't share the same manufacturers of CPU/GPU? Nintendo moved from AMD to nVidia for Switch and it couldn't play (digital) Wii U games, whereas Wii U / Wii were BC with their predecessors. I realise game format also plays a role in BC, but that wouldn't stop Switch playing digital Wii U games. I also realise that if emulating, you need more headroom between consoles than Switch has over Wii U, due to going from traditional home console to hybrid...

2) I think it's safe to say that you don't qualify as an 'average joe' when it comes to gaming :)

3) I wouldn't say display size is irrelevant, it's half the equation when it comes to pixels per degree (or whatever the actual unit is called). Most people aren't buying a larger TV/monitor so that they can sit further away and end up with the same effective resolution as their current display. Also, unless people are buying a new house to go with their bigger TV, a lot of people simply won't have the option of moving further away from it.

Even the UHS-I-limitation of the Steam Deck is good enough for 99.9% of the games (and much faster than the HDDs in PS4 Pro/Xbox One X).

If Nintendo uses a UHS-II microSD card reader, up to 300 MB/s reading speed is possible.

I inferred this from the fast load times of the Zelda demo at Gamescom.

https://rectifygaming.com/zelda-botw-nintendo-switch-2/

In the demo the load time was instantaneous, but usually it is 30 seconds.
Switch cartridges load at 60-95 MB/s.

Microsoft did it with both the Xbox 360 and the Xbox One, though it took a few years in either case.

Also Sony, but they would include past-generation hardware to do it. Not exactly a cheap proposition nowadays or even 15 years ago, when they stopped it.

Gamecube/Wii was actually designed and developed by a company called ArtX which was comprised of a heap of Silicon Graphics engineers that helped designed the Nintendo 64 GPU.
ATI purchased the company just before the Gamecube dropped.

It is functionally different to the WiiU Radeon based GPU which has full backwards compatibility with the Wii.

3DS and DS also had very different graphics capability.

Not needed.

You just need to cache the right data on internal storage, keep the rest on MicroSD, which is more than manageable these days.

Keep in mind that Microsoft had the foresight of backwards compatibility, so they kept a few things from the Xbox 360 GPU such as texture format support on the Xbox One to increase emulation efficiency, otherwise those Jaguar cores would have fallen over.

Asked & answered - thanks!

Personally I think I'd prefer Nintendo to take the risk and implement DLSS. Especially as the likel

yhood is that they'll stay with nVidia (the big 3 seem to stay with their previous gens manufacture more often than not in the recent era).

The benefits of DLSS at low res over FSR, outweigh the potential cons/risks, but that's just me. 

They showed the Matrix demo on a PC dev kit... the BOTW demo probably too.

If that's the case, the prepared demos were probably already completely in the RAM.

Hm, given that DLSS executes as a post-process, I don't think it would be a bigger issue for backwards compatibility than moving away from a dedicated Nvidia GPU would be in the first place. 

I'd assume that in that scenario Nintendo would go the software emulation route, rather than duplicate hardware. If they are software emulating, then modifying which sort of post-processes in a game to game basis or in general occur is relatively trivial. Even if they duplicated hardware they could add a wrapper to the execution step to switch out post-processing tasks. 

Hell, by then there might be dedicated post-processing chips that do the sort of stuff we're seeing now with combining multiple GPU's.

See: DLSS 3 + FSR 3 stacking. 

By the time Switch 3 releases I expect machine-learning aided graphics are going to be far ahead what we have now. Nvidia's current white papers already allude to what that future looks like. You wouldn't care about DLSS 2.1 or 3.5 in that scenario because you'll have much better upscalers and imputers that you would implement at the level of the emulator and get better performance/image quality than DLSS 2.1/3.5.

On the topic of scaling, I think games like Starfield are poor examples of scaling, if we take it from the perspective of does one's experience change much playing on different platforms. 

Basically playing low-medium on my Rog Ally as a handheld looks very similar (from a feature-set perspective) to my Series X which in turn looks very similar to my RTX 4080 desktop. It's really resolution and frame-rate that are most noticeable between these platforms. It has actually been a seamless transition to go back and forth from playing Starfield on the Rog Ally (@ variable 35-50fps), to the Series X in my living room on a 4k 60hz VRR display (@30fps), to my PC in my bedroom on a 4k 120hz VRR display (@ a variable 60-90fps.)

It is almost a meme how close Low, Medium, and Ultra are in that game. 

https://www.youtube.com/watch?v=nFa6waRs-L0

*Look at the comments, as others seem to agree.*

On the other-hand, a game like Alan Wake 2 doesn't run properly on pre-Turing/RDNA platforms due to lack of DX 12 Ultimate support. So from a "playable on x platform" perspective, Starfield scales pretty well in that you can play the game on old platforms, which is something you can't really do with Alan Wake 2. 

Having said that, I don't think a PS4 would be able to play Starfield at a stable 30fps though, in its current state, unlike the 2600k + HD7850 platform, because the game can be quite CPU heavy and the difference between a 2600k and the PS4's Jaguar in terms of IPC and overall performance is pretty steep.