I found this article on NVIDIA's work on ray-tracing-optimized GPUs interesting. The main idea is emerging hardware for ray tracing via (1) custom logic for accelerating ray-box and ray-triangle tests, and (2) support for efficiently reordering ray-tracing computations to maximize memory locality (called ray scheduling).
knguyen0811
Aside from the constraint that the number of buckets is usually less than 32, is there some heuristic that determines the optimal number buckets, using metrics such as number of primitives, size of primitives, etc.?
eliot1019
How does the bucket get computed using a centroid of an object?
I found this article on NVIDIA's work on ray-tracing-optimized GPUs interesting. The main idea is emerging hardware for ray tracing via (1) custom logic for accelerating ray-box and ray-triangle tests, and (2) support for efficiently reordering ray-tracing computations to maximize memory locality (called ray scheduling).
Aside from the constraint that the number of buckets is usually less than 32, is there some heuristic that determines the optimal number buckets, using metrics such as number of primitives, size of primitives, etc.?
How does the bucket get computed using a centroid of an object?