aeva, 25 days ago

@YoSoyFreeman With GPU ray tracing, perf degrades tremendously as rays diverge, and if you want to hit the fast path at all you need to conform to a strict subset of the functionality available. You end up with a mountain of hacks to make it workable for anything beyond trivial applications.

Using GPU ray tracing to accelerate CPU computation usually means 1-3 frames of latency for readback, so it's an impractical substitute for stuff like physics queries.

aeva, 25 days ago

@YoSoyFreeman it's also often not the fastest or even the most accurate way depending on what you're doing