Switch 2 Speculation

Nintendo and 4K? If the device outputs a 4K signal it will be for video playback alone, I mean their current box barely does 1080p and it has not been a problem for them why would they suddenly dive in to the performance wars again?
 
Look at the amount of GPU you need even for DLSS Low, that is far, far, far beyond the capabilities of a handheld platform. Now this is Nintendo who managed to very effectively pitch an SD device in the HD era (Wii) so maybe they will return to the TV box world with a new platform but I don't see a justification for it. They got the Switch by adapting Nvidia's failed ARM platform ambitions, the easiest path would be to keep it a handheld, process shrink that chip, speed bump it and call it Switch 2 with a 1080p screen. You get full back compatibility and you don't blow out your h/w R&D and dev costs for a fight you can't win versus Sony & MS.

Has anyone seen any signs of Nvidia spinning up the R&D on ARM recently? Their R&D spend seems to be all GPU focused of late
 
Good point but I haven't heard of any actual wins with Xavier which seems like a poor match for a mobile part in any case and Orin seems a ways off. Maybe it's just a Tegra x2?
 
With current upscaling capabilities of flat screen TVs they don't really need 4K gameplay. However, it would be nice if menus and the general user interface at least supported 4K.
 
With current upscaling capabilities of flat screen TVs they don't really need 4K gameplay. However, it would be nice if menus and the general user interface at least supported 4K.
But if they do native 4k for UI, how can the TV know to upscale everything else?

Btw how good is the 4k up scaler on the new shield TV? IIRC it was marketed as having great 4k upscale.
 
Aren't there a whole lot of games that render the ui separate from the game? I don't think that will be the problem.

But why bother? I'm streaming all my games from pc to my TV in 1080p and it looks fine. Same goes for the switch hooked up directly.

Not saying 4k isn't better but if you look at the more demanding switch games than 4k isn't exactly what I would call a priority.

Anyway I don't think that if Nintendo releases a new model next year it will be very different from the current one. Maybe it can output in 4k for marketing purposes but I don't see any massive changes to the SoC coming.
 
Look at the amount of GPU you need even for DLSS Low,

What do you mean by that? 4K DLSS runs in just 2.5 miliseconds on a 2060 (1.5 ms on a 2080 Ti). On a console that's probably going to be locked at 30fps == 33ms frametime, even 10-15ms for DLSS would be more than sufficient. They may even be able to run it in parallel, adding one extra frame of delay and thus have the whole 33ms for DLSS to do its thing, tho I think that would be completely pointless since Tensor flops are relatively cheap.

For reference:
2060 is around 100 TOPS (220 TOPs 2080 Ti).

Tegra Xavier does 20 TOPS @20W - Probably enough for 10ms, but too much power.

The upcoming Orin was announced as 200 TOPS @65W. Divide that by 5x with a good mix of "cutting down" and lower clocks and you'd end up with a very good 5ms DLSS runtime under 10W.

that is far, far, far beyond the capabilities of a handheld platform.

No it's not, as can be seen from the above. Achieving it would most certainly require a custom chip, because none of the ones I mentioned would be ideal, but is perfectly doable.

We can even look at it from the completely opposite direction, starting from the Ampere A100 PCIe with its 250W for 624 TOPS* and dividing,since A100 has 7 GPC enabled. So if we imagine a naive cut by 7 for a single GPC, we end up with a 35W part capable of 90 TOPS. Then move to the LP process instead of HP, lower clocks and voltages like crazy and once again over 40 TOPS for less than 10W becomes more than viable. Area should be 100mm2 for the GPU again according to naively cutting down A100. Such a part btw, based on A100's TOPS:FP32 would also be a 1.25 TFLOPS GPU vs Switch's 0.4 TFLOPS.

*1248 TOPS with sparsity, which I don't know if DLSS would benefit from, but if it does... oh boy.
 
Ok thanks for the clarification on Tensor cores, it's the silicon and energy budget that I remain sceptical because my priors on Nintendo are that they are out of the custom chip game and will prefer an off the shelf solution for the eventual successor platform to Switch that will also be a handheld or hybrid device. Tegra x1 was not the latest to market and represents Nvidia's last serious push into mobile devices, their subsequent platforms have been aimed at vehicle and other applications where power is not really a concern. There is little incentive for them to produce another mobile chip on their own given the complete lock Qualcomm has on mobile radios and the standalone tablet style SoC market never materialised.

All that being said I have painted Nintendo into a corner, I don't believe they're going to spin up a chip design team as Apple have nor do I think Qualcomm will produce a chip with a GPU that would challenge what Nvidia have or could bring to the table. So my belief is we are more likely to see an evolution of the Tegra x1 or x2 design on a better process for a 1080p target with 4K output support for whatever Switch 2 is. I also didn't think Sony or MS would go SSD, much less Turbo SSD so.... :D
 
I've been away from here so apologies if this was mentioned before, but it looks like these past weeks (months?) nvidia updated the roadmap of their jetson devkits. In that same roadmap they mention an update to the Jetson Nano called "Nano Next" coming in 2021.


alrOgei.jpg



The current Nano is using the TX1 chip, which sort of implies the Nano Next will be using a new chip that isn't Parker or Xavier (or Orin that isn't coming before 2022).

It could be that this Nano Next is simply using the 16FF Mariko chip (especially if we consider the $130 price for the board). However, I thought most of the IO and ISP hardware had been stripped down in Mariko since neither the Switch or the ShieldTV use it.


Regardless, if the Nano Next has a new chip (not Mariko) then it could mean a new chip for the Switch refresh.
 
Nice find, yeah I'd be willing to wager the Switch Pro is based on this Nano Next. It's placed on par with TX1 rather than higher but the chart seems very much segmented by target market rather than performance or any other metric so Nano Next may yet offer significant performance boosts over Nano. Looking to the future the implication appears to be that Orin will replace the mid to high end range but that Nano class hardware will continue on separate development track
 
Nice find, yeah I'd be willing to wager the Switch Pro is based on this Nano Next. It's placed on par with TX1 rather than higher but the chart seems very much segmented by target market rather than performance or any other metric so Nano Next may yet offer significant performance boosts over Nano. Looking to the future the implication appears to be that Orin will replace the mid to high end range but that Nano class hardware will continue on separate development track
Agree, “Next” certainly implies a direct lineage.
 
The most probable choice is to find a non-Nvidia SOC for future next-gen consoles, and use 7nm tegra for backward compatibility and a secondary SOC.
 
The most probable choice is to find a non-Nvidia SOC for future next-gen consoles, and use 7nm tegra for backward compatibility and a secondary SOC.

I believe the Maxwell/Pascal architecture never went smaller than 16nm FinFet. Are there any limits on how far you can shrink a processor before it breaks compatibility?
 
Back
Top