But it’s 100% not linear, right? Especially if the intercept is 0.
4.2W - 660MHz
9.3W - 1,125MHz
12.0W - 1,380MHz
Or more simplistically, 9.3w is 220% of 4.2w but the clock speed is only 170% of 660MHz.
Next jump isnt linear either: 12.0w is 130% of 9.3w but the clock speed is only 23% faster
Looking at 4.2w and 12.0w, you have nearly 3x power draw at 2x the clock speeds
All I can do is say look at the math: change in x and change in y isn’t linear.
I'll try not to get dragged into the pedantry (although I do love some good pedantry), but it's linear in that it fits a
linear function of the form y=ax+b (notice the +b, it doesn't have to intersect at 0) with an R-squared of pretty much 1. Or, to put it more simply, it's a straight line.
Regardless of how we want to define linear, though, my point doesn't change; it doesn't match the shape we would expect on a real chip, or even a reasonable estimate of a real chip. Even with just three points, this is easy enough to determine, because in the second-order polynomial form you'd expect the data to match, no set of three points will ever fit a linear function*.
* Ok, if we want to get really pedantic, there isn't going to be a precise second-order polynomial relationship between clock speed and power consumption. More accurately, power is a function of voltage, clock speed and temperature (via the Poole-Frenkel effect) split across static and dynamic power draw. A good description (with the full functional form) can be found on
this Anandtech thread, with measurements. Strictly speaking this does permit multiple data points to fit a linear function, but only if the voltage is constant between the points. I think we can safely assume that Nintendo wouldn't use the same voltage at 660MHz and 1.38GHz, so with proper DVFS we should still have something decidedly non-linear.
We know who the super VIP is? No
What company they come from? No
If they even even visited Nintendo? No
Jensen Huang
Nvidia
Yes
I mean, it's not that hard to decode. They're not going to publish a piece about Nintendo's next (obviously Nvidia-powered) hardware based on some IBM or AMD exec travelling to Japan.
Even if it's accurate, though, it doesn't really tell us much. Jensen is hardly going to be hand-delivering Drake chips on his private jet. Nintendo's a big client for Nvidia, so every so often it's worth having in-person meetings to manage the relationship and make sure everything's running smooth. Maybe this is to mark a milestone, like Drake hitting manufacturing, but it could equally just be Jensen scheduling a trip to meet with several clients/suppliers in Asia, and Nintendo is one of those he's meeting with.