Fixed MIS and Fresnel reflections. 10m 3s with 12 threads. 512-2048 samples per pixel.

100 bounces. I see no difference from the 5-bounce version.

5 bounces. Things now look correct.

2 bounces. We can see through refractive material, but inside the refractive sphere we can see a shadow that isnt visible from outside the sphere. 3-4 bounces looks similar. You can see golden light reflected from the golden sphere in the background to the ground next to it.

1 bounce. We can definitely see the diffuse colors, along with reflections and lights from emissive materials.

0 bounces. Notice that we basically see black for non-emissive hits, and the emission color otherwise.

Samples image for my 6400-51200 samples per pixel render. Its interesting that the refractive sphere converges so smoothly compared to everything else.

Check this out! Im not sure why the walls seem to have spokes artifacts for the light, but its kind of cool. 24m 18s on an RTX 3060. Between 6400 and 51200 samples per pixel, 5 bounces per path.

The brightness is much better. Im seeing some interesting ray-like artifacts from the light on the walls. I need to add glossy sampling to the path tracer. My implementation is far slower than I think it should be - 2m 51s to render this with 12 threads, 8 anti-aliasing samples per pixel, and 200 path tracing samples per anti-aliasing sample, and 6 path-tracing bounces.

Attempt 1. This is WAY too bright. The shadows and caustics look good, though.