Is stochastic raytracing inherently cache-unfriendly?

Question

Specifically in the context of a real-time raytracer where view updates are frequent?

The obvious answer would seem to be "yes" and yet I wonder if any methods have been found to accelerate Monte Carlo methods given their usefulness.

Answer 1

If you mean by stochastic stochasting sampling (for simulating effects like DOF or motion blur), the answer is probably yes. Two sample rays for a same pixel could lead to two very different paths in your acceleration structure, leading to potential cache misses.

One of the best way to accelerate this is simply not to use raytracing for primary rays but rasterization and use stochastic sampling of your polygons (check for Reyes rendering). That's what softwares like Pixar's RenderMan® do for instance.