I'll amend your comment there because he's also basically talking the historic of NVdia trying to buy ARM which are the ones producing the mobile chips we see in the Switch. 

They don't have the "copyrights" for using the Tegra line anymore which lead him to believe the Switch 2 to be boasting a Xavier chip from what he allegedly saw. 

Indeed, beware of anything that runs on the internet because the only true answers we'll get are the ones we'll have once we actually see the Switch 2 for real.

Yeah i didnt know you meant at that low of res and thought you where talking about DLSS in general.

Yeah and that was my bad.  My post wasn't clear.  

In this podcast from Nate the Hate, he says that the Switch 2 was able to run a demo of Breath of the Wild at 4k 60FPS, with instant loading times. It was using DLSS 3.5. He also mentions that The Matrix Awakens demo was running on it. using advanced Ray Tracing. In the end, he says that "March 2024" was heard a couple of times, but without any context, so we don't know if it's the release date, the release window, or the reveal date.

The improvements in power consumption from smaller nodes are, IPC improvements and different clock-voltage curves aside, a direct function of the physical size of their transistors.

The scaling between TSMC's 16FF and N5 is 88 x 70 nm / 51 x 28 nm ) = 4.36 times. For actual GPUs in real-world scenarios, you can see the improvement was more like ~ 3 times between a GeForce 10 and a GeForce 40:

So yeah, the overall improvement should be around that ballpark. It has to.

Mind that Ampere GPUs can also dual-issue FP32 instructions if there's no simultaneous INT32 going on. So it could claim to be a ~ 4TF console but perform closer to a PS4 than a Series S (without DLSS).

It is important to remember that comparing Tflops across architectures is a dangerous game.

Tflops, as a real-world metric, make sense for FLOP-heavy tasks like scientific computing (i.e, you're basically solving matrix differential equations, for example.) 

Gaming is a mixed load where int operations happen for about every two floating point ones.

There is also the complication that RT cores and tensor cores provide, in that they accelerate different specific types of FLOPs (ray-tracing, DLSS, etc.) Hence NVidia tries to bring in metrics like "RT TFLOPS" and "Tensor TFLOPS" to accommodate.

The Switch 1's chip was Maxwell, not Pascal. You'd want to compare the 4080 to a 980. That is more of a 5 times difference. 

The X1 in the Switch was also originally 20nm. The die-shrink to 16nm happened with later revisions, with a battery life improvement. 

16nm 🤔

So we're still in the dark. Some say the opposite than others.

Edit - Sorry I didn't see this article : https://www.vgchartz.com/article/458491/nintendo-briefed-activision-blizzard-on-switch-2-last-year/

Sounds like my PS4 raster prediction is gaining some evidence