Have you worked with RTXGI in production cases? I’ve experimented with it somewhat (I was mainly interested in trying out NVRTX’s caustics features), and I generally found it to be more prone to light-leaking then lumen. Not to mention the need to set up manual probe volumes.
Also, doesn’t RTXGI suffer from the same BVH complexity problems that bring HW lumen with overlapping messages to a halt? Just curious.