The first time I found out the Wii U existed was in 2015 when I saw it on display in an electronics store. It was hooked up to a TV and people could try out Mario Kart 8. I took a glance at it and thought: "What in the world is this?" The gamepad confused me, so I ended up just ignoring it and kept walking.

It took until the Switch released that I learned what this weird looking thing even was. Only when Splatoon 2 released I found out that that franchise existed.

Mind you I was a kid and barely ever used the internet.

The problems are:
- if dock will have the capability to run with beefier specs, it could probably run the game on itself. Just transfer memory, states, etc to dock and continue run there. But would change the point of a hibrid system.
- bandwidth. APU have their components engineered to be at right spot to deliver the maximum speed. Involving ports in other piece of hardware would make it decrease the bandwidth or a nonusual arrangement on plugging the hardware pieces.
- there would be plenty of issues when unplugging. Would be harder for developers to guarantee that any moment can be switched.

I'd rule out any increase in memory or any pre-rendering processing.
I think they have only 2 options:
- keep as it is right now. Just overclock.
- dock handles post rendering processing. Receives rendered images and apply AI super resolution. But it may also be the first case, where if the dock can perform these gpu intense tasks, it could just run the games itself.

What I think they should do:

A more modular approach, where you choose among parts. The mais tablet would have two different specs to choose.
The cheapest would have all the memory and pre-rendering gpu for all modern games for good handheld resolution.
When docked, it outputs until 1080p.

The more expensive would only include extra tensor cores for post processing high resolution(DLSS), and a screen with higher resolution (and maybe oled, higher storage). Would output until 4k when docked.
When hh, you could choose to use the extra cores or save battery.

I think in this way we will have a affordable entry point, while also feeling next gen for those who want. Developers focus on cheaper version, don't need to care about 4k, let the AI do it.

And also they can market it well, call it Nintendo Module, add extra modules for controllers and have different controllers for different gameplay, call it innovation, market on modular design where you can choose what you want, while the minimum is just a next gen switch.

All of Nvidia's modern chips have tensor cores, including Orin, you'd probably have to redesign the entire architecture if you wanted to strip them out.

Besides, DLSS is something that the developers will definitely need to care about and get to implement since the upscaling model needs to be extensively trained with each game's assets.

That's assuming Nintendo will mandate that every game supports it, of course (which I hope they will).

USB is high latency. It's crap for multi-GPU. It's also got a polling rate of 125hz verses interrupt.

USB 4.0 V2's max bandwidth is only 120Gbps (14.4GB/s) but those are asymmetrical speeds it is limited to 40Gbps in the other direction (5GB/s), SLI/Crossfire requires communication in both directions... And that is before we include encoding/decoding overheads of the protocol which lowers even those numbers.

Base USB 4.0 is 1.2GB/s. - Not enough for multi-GPU.

SO while you do have external GPU's via USB 4, they tend to come with a massive corresponding hit to performance, frame pacing issues, micro-stutter and more issues.

Plus PCI-E 4.0 x8 levels of bandwidth is starting to become a limitation in terms of GPU bandwidth.

This is also Nintendo, they aren't going with the highest spec, most expensive, most power hungry implementation of USB 4.0. That would be silly, they might as well just invest that extra controller silicon space into a beefier GPU and be done with.

This is the traditional "technically you can do that, but is it worth it?" debate...

At the moment the dock is a piece of plastic and probably costs $5. If you want to put hardware in it, it will cost more. Then trade-off putting better hardware in the Switch itself vs the dock.

The total price of the thing is going to be $300. So inevitably you make the dock better by making the portable Switch worse.

How would a Thunderbolt 4 port fare in such cases?

As much as I like the possibilities of modularity, I just don't see Nintendo doing it. their more "simple to understand" devices are usually their most successful ones as well. But this thread is exactly for that: abstracting what Nintendo generally does and thinking of possibilities, so this is a very interesting one!

Equivalent to PCI-E 3.0 4x or about 5GB/s Aka 40Gbps. It's simply not enough.

But it does have a PCI-E mode.

Thunderbolt 5.0 will double that, but I would argue that is still not enough.

You do get to a point where you invest so much time and energy in a connection interface, that it would actually just be better to have a larger SoC.