They're doing it the smart way this time around. Instead of random LOD's pop ins or excess polygon count that you can't see on screen, you can shoot for a polygon/frame for more predictable outcomes and sustainable. All platforms should benefit from that, but will scale according to how fast they can stream things. Actually 20 million polygons per frame is quite excessive, it's above native 6K theoretically, but provides much cleaner final image closer to real life footage.
You would need at least 1 polygon per frame to mimic similar results but with less, meaning 8.5 million instead of 20 million (4K is 8.5 million).
That demo was crunching 4.7 polygons per frame per pixel (1440p, ~4.2 megapixels). Anti-aliasing is out of the window as well, I guess.