Carzy Zarx’s PC Gaming Emporium - Catch Up on All the Latest PC Gaming Related News

[Pemalite]

JEMC said:

caffeinade said:

Nintendo is pretty cheap, they're not going to pay for all that silicon.
Physical size matters in mobile devices.
A base Tegra X1 has a TDP of 15 watts, not 20, meaning you'd have to cut it down even further to reach stock X1 power consumption.
The Switch uses an underclocked X1.

Xavier wasn't built for gaming.
It will only be decent at gaming by brute force.

Nintendo would be better served with an SoC that is custom designed for their needs.
And have a GPU that is balanced for the task at hand.
They'd probably want to find the optimal frequency for their architecture, 512 core will probably be too much to hit the sweet spot.
Also we don't know how Volta compares to their gaming architecture, in games (if they are indeed separate things).

Yes, Nintendo is pretty cheap, which is why they may go for an off the shelf product rather than paying tens or hundreds of millions to Nvidia to develop a custom SoC. And also because it's Nintendo, them going with Xavier for the Switch 2 is likely as by the time they launch it, it will be "old" and cheaper. Heck, by then it may also be available at 7nm, making it smaller and less power hungry too!

But well, Nintendo being Ninendo, they could go with an X2 for the Switch 2 and call it a day.

Pemalite said:

Nintendo Switch's SoC is 232mm2.
Xavier is about 300m2.

That is roughly a 30% chip size increase.
However... 12nm is very high yielding anyway, so that increase in chip size should be economical for Nintendo.

Xavier most certainly was built for gaming, it just so happens it is good at other tasks as well.

With that in mind... I think the better idea would be to take Tegra X2, port it over to the 12nm process and dial up the clock rates, Pascal isn't a big deviation from Maxwell from a feature set perspective... But can hit higher clocks for better performance. (Plus improved Delta Colour Compression for a little more bandwidth.)

Wouldn't a gaming device like the Switch benefit more from having twice the shader cores instead that a higher frequency?

Well. It depends...
Going wide (Twice the shader cores) doesn't always mean it will be more efficient than taking a design with half the shader cores but with twice the clock, the fabrication/transistors used and the energy they consume have a big influence to that end as well.
So it's hard to pin down a definitive answer for that.

Pascal tends to clock higher than Maxwell because of the newer fabrication process as the newer, smaller transistors could clock higher at lower voltages... Not to mention the various architectural tweaks nVidia employed to obtain higher clockrates and reduce leakage.

But one thing more shader cores generally does bring is a larger SoC which typically means higher costs. (And that isn't even black and white anymore due to other factors in fabbing chips these days.)