Anyone watched/watching GTC? Watching part 2 there was something very off about there little game demo. No one at NVIDIA must have played with marbles, as the shadow was completely solid Watch from 10:45 it's just faked EDIT: even the thumbnail shows it shouldn't have a solid shadow!!!
Even with RTX raytracing... reflections are easy, caustics and refractions are not... so some fakery will happen.
I think the line "physically based materials, obeys the laws of physics" after the clip needs a rather large asterisk. lol
Refractive caustics are hard, even for unbiased path tracing. Just throwing more 'ray bounces' at the problem (already an exponential increase in compute difficulty for every bounce added) can only approximate the solution, full refraction simulation with variable refractivity gradients (not just homogenous bulk inter-layer refraction) needs a whole new ray physics model.
I recall having to render a scene with very visible causitcs [ye stereotypical light through prism] and not only just having caustics turned on send the render time through the roof, but each frame had to be rendered multiple times, each with a marginally different IOR, and the composited together to get something that looked even vaguely realistic bc the renderer couldn't do light dispersion. [and even then, the long you looked at the more obvious it was that it was a 3d render]
