> rendered with an approximately 2 GB geometry pool, in 26ms when using ray tracing and 16ms when using rasterization, on NVIDIA GeForce 3050. Not bad for a GPU that draws all its power from the motherboard without needing a separate power cable!
This is very impressive, but I still think we're heading in a really scary direction with this combination of art & technology for gaming use cases. Memory bandwidth is finite. Giving an art team permission to use assets of arbitrary complexity is a dangerous precedent. It is impossible to put this kind of cat back into the bag once you start having trouble with it on lower-end platforms.
Resorting to TAA is inevitable if you lean on Nanite. All (efficient) intraframe anti-aliasing techniques fall apart with sub-pixel triangles. SSAA defeats the core principle of Nanite and MSAA would be potentially be even worse in the most degenerate cases.
