New Switch Tegra X1 "Mariko" Chip Details

[Pemalite]

curl-6 said:

Pemalite said:

I concur. 12nm (Based on 16nm which in turn is based on 20nm.) would be the most ideal fit... But that is still a dramatic improvement over the archaic 20nm process.

Very interesting that Nintendo isn't just simply dropping the Pascal derived Tegra in... And nVidia decided it was worth the effort to engineer/respin a new SoC for Switch and Shield.

No reason why we can't have both performance gains and battery life gains though.

I can't help but feel like it'll be 16nm simply cos Nintendo always seems to play it conservative when it comes to tech stuff.

As for not using X2, could going from Maxwell to Pascal conceivably cause any compatibility issues, could existing Switch games perhaps use some Maxwell-specific code? Maybe it was simply cheaper this way and they don't wanna fork out the cash for the X2.

There isn't really any significant changes between Maxwell and Pascal from a developer point of view that I can think of.
Nor do we know if the new SoC is based on Maxwell anyway.... So it's an extremely interesting situation either way.

16nm is probably not as cost efficient as TSMC 12nm right now which seems to be the "sweet spot" in terms of fabrication, hence why nVidia is using it for Turing... Plus because 12nm is based upon 16nm, it's really really easy to port designs over to 12nm anyway.

I would assume price will play a factor into it, hence why nVidia is likely piggy-backing on Nintendo's potential scale-of-economies with Switch to drop it into Shield.
I would think it has some advantages over the Pascal based Tegra X2 though, otherwise there would be zero point in nVidia using it over the Tegra X2 in their  Shield devices.