I remember some people laughing at the 8-10 nm chip on the Switch 2 when the specs were leaked, saying that the SteamDeck 6 nm chip was way more power efficient and the SW2 was going to have an ancient power hungry chip… What happened?
What happened is they don't know how semiconductors work and are used to node % gain metrics seen for marketing purposes which are often just metrics for the performance of a gate only and at the highest clocks.
Nvidia has never used the full density of a node be it Samsung or TSMC, they prefer low density thermals and stability, nor can you even approach theoretical densities most of the time with a reasonable complex architecture but Nvidia is on the lower side.
Switch 2 is large-slow over a narrow-fast design. Complete opposite of say an Apple chipset with small silicon and high clocks or in some similar sense an AMD portable APU found on Steam deck or other PC handhelds.
Voltage drop is DIRECTLY related to smaller nodes where resistance of interconnects within a chip increases. High densities + fast switching speeds (clocks), they're culprits in creating challenges with voltage drops and this is where small nodes are preferable.
So because Nvidia engineers are not newbs (surprised pikachu), they knew already from their Orin AGX platform made for stability, harsh environment, low power, low clocks that it would NOT be impacted as much on samsung 8nm over say paying super expensive prices for TSMC 4nm or something. Even going TSMC 5nm I bet it has not much impact on that architecture. There's leeway in the future for a node reduction and a boost in clocks though for a Pro version. If clocks remain the same, it would have less impacts.
Do the SW2 doubters have at least one win in their technical analysis here?
Nope. I warned multiple times about exaggerating the impacts of 8nm. But what do I know, I only studied semiconductors in electrical engineering, but more importantly I'm not even 1/1000th remotely close to the understanding that an Nvidia designer would have to be assigned on an APU and finding the best node for mass production of a potential >100M units device where yields also have big impacts if you're too aggressive. The chipset ended up way more efficient than I thought it would. I kind of knew that low clock chipsets were less impacted but right now its a slaughter. Bring down an AMD portable to 9 Watts to see what happens.
Thinking that Nvidia would drop the ball here is laughable. Just like peoples thinking it won't have enough TOPS for DLSS, or that there's RT hardware there but it'll be so weak that it's not gonna be used other than shadows if even that, dead silicon as if a designer would put ML and RT units on a custom APU if it was not usable.. Peoples calling it a Switch pro...

I've seen it all. All those old argument pillars I've fought are now dust.