AMD Cuts Ray-Tracing Memory Use by 97% with New Cage Tech
A new AMD technique slashes VRAM requirements from 80GB to 1.7GB for complex scenes, though it limits individual vertex control.
Key points
- AMD's tetrahedral cages reduce ray-tracing VRAM usage from 80GB to 1.7GB for dense scenes.
- The technique cuts BVH update time from 300ms to 3.3ms per frame.
- Developers lose individual vertex control, making it best for physics-based environments like forests.
AMD has presented a new ray-tracing method that dramatically lowers the hardware demands of rendering complex, animated environments. By using a technique called tetrahedral cages, the company demonstrated a dense forest scene running at over 60 frames per second on a Radeon RX 9070 XT. This performance is achieved with a fraction of the memory usage required by traditional methods.
According to reporting from TweakTown, the system reduces the number of ray-traced triangles per frame from 2.8 billion to roughly 500 million. More significantly, it cuts video memory requirements from 80GB down to just 1.7GB. The update time for the scene's geometry structure also drops from 300 milliseconds to 3.3 milliseconds, enabling real-time animation without sacrificing frame rates.
Static cages reduce memory overhead
Traditional ray tracing requires constant updates to the scene's bounding volume hierarchy every time an object moves. This process is computationally expensive and memory-intensive. AMD's approach wraps dense meshes, such as thousands of grass blades, in a deformable cage. This cage allows the underlying hierarchy structure to remain static and reusable across frames.
Instead of recalculating the entire structure, the system deforms the cage at runtime. Rays are first compared against this static reference frame before intersecting with the detailed triangles inside. This means the animation cost scales only with the complexity of the cage itself, rather than the massive number of individual triangles contained within it.
Trade-offs for animation control
This efficiency comes with a significant limitation. Developers lose the ability to animate every individual vertex in the scene. The technology is purpose-built for environments where individual control is less critical, such as forests or fields of grass. In these cases, physics-based effects like wind are typically used to create movement rather than manual keyframe animation for each blade.
This makes the technique ideal for dense, organic environments but less suitable for characters or objects requiring precise, independent motion. The trade-off is a loss of granular artistic control in exchange for massive performance gains. It effectively shifts the workload from geometric updates to simpler cage deformations.
Research phase before game adoption
The technology is currently in the research phase and is not yet available in game engines. AMD confirms compatibility with Microsoft DirectX Raytracing, suggesting potential future integration into mainstream titles. It could appear in the current RDNA 4 generation or future console hardware, provided developers adopt the workflow.
For this to become common, engine developers must implement the new pipeline and artists must adapt their workflows to use cage-based deformation. While the performance benefits are substantial, the shift in animation technique requires industry-wide adoption. It is a promising step toward more realistic environments, but it remains a specialized tool for specific types of scenes.
