AMD presents method to reduce memory usage in ray tracing

In a research paper, AMD specialists demonstrated the new method using a scene with roughly 25,000 animated plants as an example. Under the traditional approach, processing such objects would require up to 80 GB of video memory for bounding volume hierarchy (BVH) acceleration structures, while updating them took more than 300 ms per frame. Using tetrahedral cages, memory consumption dropped to 1.7 GB (roughly 47–50 times less), and the update time was reduced to 3.3 ms.

The core idea of the technology is that detailed geometry is placed inside a simplified deformable cage. The 3D models themselves remain static, while only the outer shell is animated. During tests on a Radeon RX 9070 XT graphics card, a scene ray tracing around 500 million triangles ran at over 60 frames per second at a resolution of 1920 × 1080 pixels.

The company's research received the Wolfgang Straßer Best Paper Award at the High-Performance Graphics 2026 conference. The development is currently at the research stage and has not yet been integrated into finished commercial games. Specialists noted that the technology is optimal for foliage, grass, or background characters, but is not suitable for detailed facial animation due to the inability to control individual vertices.