Ray Tracing Cores for General-Purpose Computing: A Literature Review

A team of researchers led by Enzo Meneses has published a comprehensive literature review analyzing the emerging field of repurposing ray tracing (RT) cores for general-purpose computing. By systematically examining 35 research articles, the review identifies patterns in how these specialized GPU components—designed for realistic lighting in games—can be used to solve computational problems by reformulating them as geometric queries. The analysis reveals that problems like nearest neighbor search, common in AI and databases, benefit most significantly, with some studies reporting speedups of up to 200x compared to traditional methods on the same hardware.

The key insight is that RT cores excel at tasks involving tree traversal where large portions of the search space can be heuristically discarded, avoiding unnecessary work. The review categorizes 32 distinct problems successfully accelerated, most falling into physics simulations and geometric queries, but also notes potential applications in AI. A critical technical finding is that workloads generating many short rays are more efficient than those with a few long rays. This work serves as a practical guide for engineers and researchers, providing clear criteria to evaluate whether a given computational problem is a good candidate for RT core acceleration, potentially unlocking massive performance gains from hardware already present in modern GPUs.