Efficient Ray Tracing of Subdivision Surfaces using Tessellation Caching

Figure 1: Example subdivision surface scenes rendered with diffuse path tracing (up to 8 bounces, 7 − 12 secondary rays/primary ray). Right: the Courtyard scene (66K patches after feature-adaptive subdivision) is adaptively-tessellated into 1.4M triangles from scratch per frame, and ray traced with over 90M rays per second (including shading) on a high-end Intel R©Xeon R © processor system using our efficient lazy-build caching scheme. Left: four Barbarians embedded in the Sponza Atrium scene (426K patches) and adaptively-tessellated into 11M triangles are ray traced with 40M rays per second. A 60MB lazy-build cache allows for rendering this scene with over 91 % of the performance of an unbounded memory cache. Compared to ray tracing a pre-tessellated version, the memory consumption is reduced by 6−7×. A common way to ray trace subdivision surfaces is by construct-ing and traversing spatial hierarchies on top of tessellated input primitives. Unfortunately, tessellating surfaces requires a substan-tial amount of memory storage, and involves significant construc-tion and memory I/O costs. In this paper, we propose a lazy-build caching scheme to efficiently handle these problems while also ex-ploiting the capabilities of today’s many-core architectures. To this end, we lazily tessellate patches only when necessary, and utiliz