For reference GlobalFoundries (who AMD left because they weren't willing to adopt EUV which is needed for cutting edge geometries) goes up to 12nm, so that means there is still room for Pis to improve before hitting the expensive stuff.

The Pi is great! Just saying, running a Pi with an object detector on batteries draws around 9W. That's big batteries and solar or wildlife use cases. But think how cool it would be if it could run on 2W.

Having said that. The Jetson runs at around 7W doing that and a lot faster inference. Just a lot more expense. Likely the Jetson will eventually be something like 2nm. So maybe we see around 4W at that time running an object detector. And of course the Jetson has the memory and processor together on it's modules so everything can benefit. Just dropping the power on the Pi for the processor leaves that out.

But thanks for the links

Jetson Nano was manufactured using the 20nm node as far as I can see[1].

Just goes to show manufacturing node isn't everything. There's a lot of various optimizations one can do depending on targets.

Guess we'll just have to see what the future brings.

[1]: https://en.wikipedia.org/wiki/Tegra#Tegra_X1

The latest orin modules are 4nm I'm pretty sure.

Samsung 8nm for Orin