I agree with you that I can't make 8nm work out. But I am also not an electrical engineer.
What I ask myself is, can i get close enough that I think it's
possible that Nvidia figured out how to close the gap using other power saving tech.
My personal thinking - and maybe you have a better route - is that I can't figure out Orin's true overhead by only comparing one Orin variant to another The Snapdragon 480+ is manufactured on 8nm. It's got two A76 P-cores, plus 4 E-cores, an Adrendo 619 GPU, and a LPDDR4 memory clock.
When you run a high end benchmark on a device with a mix of P and E cores, the benchmark tends to run exclusively on the P cores, because that's designed for peak performance. I've seen benchmarks of the 480+ showing the p cores, in a passively cooled device, holding 1.5-2.0Ghz on 15 minute long benchmarks, without thermal throttling.
The 480+ has a 3W TDP. The Orin Nano, with only two cores enabled, with the EMC running at the LPDDR4 speed, with 1.5GHz CPU clock, running under "medium" load, the GPU disabled, is running 6.7W.
The 480+ can't disable its E-cores. It can't disable its GPU. The P-Cores are a generation older than Orin's. This is a test designed to make the 480+ look power inefficient and it still draws 3.7W less power than Orin. So I suspect there is at least 4W, if not 5W, of savings to be had.
That's still not enough to make 8nm viable to me. But for my brain - and this is just me - it also tells me I don't have enough data about 8nm performance to be confident in my assessment. That's why I put the odds so low.