AMD’s New Ray Tracing Technology Officially Takes Flight
AMD is working on new methods to lighten the computational side of ray tracing instead of relying solely on more powerful hardware. The company’s research project, called “tetrahedral cages,” demonstrates its ability to work much more efficiently than existing methods, especially in scenes with thousands of moving plants or similar dense geometries. A technical demo performed on the Radeon RX 9070 XT clearly showed the method’s potential.
AMD Reduces Memory Usage with New Ray Tracing Method
The technology is based on placing detailed models inside deformable tetrahedral cages instead of constantly updating them. This way, the BVH structure used in ray tracing calculations is not recreated with every small movement and can be largely reused in a stable manner. The animation cost also becomes dependent on the complexity of the cage rather than the number of triangles inside.
AMD’s demo, featuring approximately 25,000 animated plants, shows a significant difference. The traditional method requires around 2.8 billion ray tracing triangles and approximately 80 GB of VRAM per scene. With the new method, the number of triangles drops to around 500 million, while memory usage falls to only 1.7 GB.
The results regarding BVH updates are similarly impressive. Update processes, which take approximately 300 ms per frame, are reduced to 3.3 ms with the new method. The demo runs at over 60 FPS at native 1080p resolution on a Radeon RX 9070 XT.
Of course, the system has some limitations. Because detailed animation control is lost on each vertex, the technology is primarily aimed at scenes like forests, grasslands, and dense vegetation. AMD’s ray tracing technology is still in the research phase, but its compatibility with DirectX Raytracing holds significant potential for future integration into game engines and existing RDNA 4 hardware.
Read More:
