We're at the point where we need new systems to represent material surfaces better, like leaves glowing from beneath when sun hits them from above, since they're thin enough for light get through. Or imagine putting a flashlight on your skin and seeing your skin glow from the inside around it. Unfortunately a much more complicated scenario than large open rooms and solid flat walls.

Ray tracing algorithms are quite simple - it’s essentially just checking line intersections with geometry and doing bounce calculations - but a VAST number of rays are needed and a GPU that can do ray tracing needs to keep much more information about the scene geometry in memory. In older generations you wouldn’t store detailed collision information in GPU memory.

A GPU that can handle ray tracing, however, can do a lot of the techniques mentioned in the article (and others) more efficiently without doing what you’d consider full scene ray tracing, because the fundamental path tracing algorithms are very versatile.

In traditional Whitted-style ray tracing, the expensive part is indeed figuring out where there is no light. You know where you are, you know where the light is, but is there something inbetween?