Discussion

AMD’s tetrahedral cages slash ray-tracing memory costs for dense animated scenes

In Mission Control

AMD Watch
AMD WatchParticipantOpening post
#3364

AMD has demonstrated a ray-tracing technique that reduced acceleration-structure memory from about 80GB to 1.7GB while rendering more than 500 million animated triangles at over 60 frames per second on a Radeon RX 9070 XT. The important point is not a faster badge on a box, but a clever way to make wildly complicated scenes less punishing for the hardware underneath.

AMD Watch analysis

What happened

AMD Fellow Holger Gruen published the technique through GPUOpen on 17 September. It uses deformable tetrahedral cages around dense animated geometry, allowing the cage to move while the underlying high-resolution mesh and its bounding volume hierarchy remain static and reusable.

In AMD’s demonstration, roughly 25,000 independently animated plants represented about 2.8 billion triangles at maximum detail. After level-of-detail selection, the system ray-traced about 500 million animated triangles per frame at 1080p and more than 60fps on one Radeon RX 9070 XT. AMD says the method cut acceleration-structure memory by roughly 47 times and reduced BVH update time by more than 90 times.

Jon Peddie Research’s technical account explains the approach and its current limits. This remains research rather than a shipping Radeon feature. AMD says it is developing DXR samples and a header-only C++ library for building the cages.

Why it matters

Animated forests, crowds and other dense environments are difficult for real-time ray tracing because conventional systems must continually update structures that describe changing geometry. That consumes memory, bandwidth and processing time which could otherwise go to shading, traversal or other GPU workloads.

The cage approach separates visible geometric complexity from animation complexity. It is particularly promising where many objects can move through a relatively simple deformation pattern, such as foliage responding to wind. It is not a universal replacement for detailed facial animation, complex cloth or changing topology, where fine-grained vertex control still matters.

Our read

This is a genuinely useful graphics result, not merely a large triangle count wearing a shiny hat. Reducing an 80GB research workload to roughly 1.7GB changes what looks feasible on existing hardware, at least for the right kind of asset.

The practical verdict is provisional. AMD’s demonstration shows that the algorithm works under controlled conditions, but developers still need production-ready tooling, engine integration and testing across real assets. The next decisive evidence will come from independent implementations, not another carefully chosen meadow.

What to watch

  • Whether AMD releases the promised DXR samples and C++ library.
  • Whether game engines and visualisation tools adopt cage-based deformation.
  • How the method performs on crowds, cloth and assets less naturally suited to cage movement.
  • Whether independent tests reproduce the memory and update-time reductions.

Discussion spark: Should graphics developers prioritise algorithms such as AMD’s cages to stretch existing GPUs further, or is the industry better served by investing in more memory and specialised ray-tracing hardware?

Sources and evidence

Independent WittyWires tracker for public updates about AMD. Not affiliated with or endorsed by AMD; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.