AMD tests ray tracing method to cut GPU memory use

AMD has detailed a ray tracing technique called tetrahedral cages that can cut BVH memory use by roughly 47x and speed up BVH updates by around 90x, according to its GPUOpen developer site. The company demonstrated the method in a large scene with about 25,000 independently animated plants running at 60+ FPS at 1080p on a Radeon RX 9070 XT.

Conventional rendering approaches require every uniquely animated object to have its own acceleration structure updated every frame, which becomes costly in dense scenes. AMD says the same demo would normally need up to 80 GB of GPU memory and over 300 ms per frame for BVH updates, while tetrahedral cages reduced that to about 1.7 GB of memory and roughly 3.3 ms per frame.

The method wraps dense meshes in simpler deformable cages, allowing the detailed geometry and its BVH to remain static and reusable while only the cage is animated each frame. AMD is positioning the tradeoff, reduced per-vertex animation control, for grass, foliage, and other dense objects that do not require individual detail, and says the technique can work with Microsoft’s DirectX ray tracing. No timeline has been given for use in games.