Ray tracing massive amounts of animated geometry using tetrahedral cages

For anyone who – like me – was hoping for a video to see what it looks like:

what is it about raytraced animation that makes it more expensive to update the BVH mesh vertices every frame, compared to old-school bone animation where you send the info to the vertex shader every frame?

For anyone who – like me – was hoping for a video to see what it looks like:

https://youtu.be/h2V8YyfKCVs?t=4421

It’s not always more expensive, but generally speaking for games and assuming some simplistic & straightforward approaches on both sides…

With raster, you don’t necessarily have a BVH to update, you evaluate your rig, e.g., bone animation, which means re-calculating all the vertices in your animated object/character, and then drawing the resulting mesh.

With ray tracing, you have to evaluate the rig (do everything above) and then also update the BVH, so the BVH update is extra. It’s also sometimes expensive because if you rebuild a BVH from scratch, that means sorting the triangles, sometimes multiple times. You can sometimes get away with a BVH “refit” for animated stuff, which is faster, but has tradeoffs. Even in the best case, you still have to read all the animated verts a second time after computing them, which costs memory bandwidth.

Sometimes games with raster engines do have a BVH, but it’s likely to be a small BVH over objects in the scene, whereas ray tracing usually builds a large BVH over triangles and is a much bulkier acceleration structure.

BTW, the linked paper here saves on both rig evaluation and BVH rebuild - in other words, this would help save time even if you weren’t ray tracing. This is essentially building a lattice rig so you don’t have to do the bone animation. Assuming the lattice rig has fewer verts than the bone rig, you save on rig evaluation time. It’s also meant to build a smaller BVH - you need a BVH over the lattice cells, and the BVH inside each lattice cell can stay static and contain multiple triangles. Only the lattice moves, and you only need to update the lattice BVH, so the ratio of lattice cells to triangles in the lattice gives you the approximate BVH build speedup.

Massively simplifying here but when doing one rasterization pass you process a flat list of triangles, in order, once. When raytracing a camera view you query a database of triangles, at least once for each pixel. To avoid scanning the whole list of triangles for every pixel, the database needs a fancy 3D spatial index whose structure depends on the exact position of each vertex. When you modify the position of vertices you have to fix up or rebuild the fancy spatial index too.

> compared to old-school bone animation where you send the info to the vertex shader every frame?

If I understand correctly this really isn't a thing anymore. Instead most modern engines copy the original vertices to a temp or skinned vertex buffer with a compute shader and then use normal non-skinned rendering techniques to display them. This means you don't need 2 copies of every shader, one with skinning and one without so less combinations.

That doesn't change the other answers to your question.

To make ray tracing practical, you need to update the ray-tracing accelerating structures when you animate.

it's a global visibility problem, unlike rasterization which is a mostly localized problem.

apart from all the updates and acceleration structures you have to do, you are also battling cache coherency during ray traversal, or as the saying goes: Primary rays cache, secondary trash.

It’s not always more expensive, but generally speaking for games and assuming some simplistic & straightforward approaches on both sides…

To make ray tracing practical, you need to update the ray-tracing accelerating structures when you animate.

it's a global visibility problem, unlike rasterization which is a mostly localized problem.

apart from all the updates and acceleration structures you have to do, you are also battling cache coherency during ray traversal, or as the saying goes: Primary rays cache, secondary trash.

> compared to old-school bone animation where you send the info to the vertex shader every frame?

That doesn't change the other answers to your question.

The animation scaling problem

Ray tracing has become a practical part of real-time graphics, but animated geometry still presents a difficult scaling problem. The paper Ray Tracing Massive Amounts of Animated Geometry by Gruen et al., winner of the third-place Wolfgang Straßer Award (Best Paper) at High-Performance Graphics 2026, tackles a common production case: scenes with many complex objects that all deform differently. In a conventional ray-tracing pipeline, every uniquely animated mesh may require both its vertices and its acceleration structure to be updated each frame. That computational cost grows with triangle density, so geometrically dense foliage, grass, crowds, or creatures can quickly consume the frame budget. As each unique animated object may need to store its own set of bounding volume hierarchy, memory consumption may become prohibitively high.

Ray Tracing Massive Amounts Animated Geometry 01 Figure 1. Animated scene with roughly 585 million animated triangles @ 60 frames per second (AMD Radeon™ RX 9070 XT GPU at 1080p)

Decoupling animation and triangles

The central idea of the paper is to decouple animation cost from triangle count. Instead of applying animation directly to every triangle, the method builds a low-resolution tetrahedral cage around the original, non-animated mesh. During preprocessing, the mesh is split into small, disjoint pieces associated with the tetrahedra. Static mini-BLASes are then built for these pieces once and reused. At runtime, only the cage is animated. Rays that enter an animated tetrahedron are transformed into the corresponding rest-pose space, where they can intersect the static geometry. The resulting motion is an approximation: vertex movement is represented by a piecewise-linear deformation induced by the cage, rather than by a smooth global bend or by full per-vertex animation.

This makes the approach especially attractive for massive scenes where many copies of the same asset are animated in unique ways. Each unique deformation needs its own animated cage, but it can share the same rest-pose mini-meshes and mini-BLASes. The paper demonstrates this with trees, grass patches, and frogs, and shows a combined scene of roughly 585 million animated triangles rendered at 60 frames per second on an AMD Radeon RX 9070 XT graphics card at 1080p. The key message is not that tetrahedral cages replace every animation technique, but that they provide an additional new low-cost option when acceleration-structure updates, animation time, or memory footprint become the bottleneck. The paper also sits alongside similar independent work by Luton and Tricard (2026), which explores tetrahedral indirection but uses multiple hardware rays and no clipping step. Please note that tetrahedron-cage driven animations can be combined with the Microsoft DirectX® Raytracing (DXR) Functional Spec, Part 2 – e.g. partitioned top-level acceleration structures – and can also coexist with cluster level acceleration structures.

Ray Tracing Massive Amounts Animated Geometry 02 Figure 2. High-level workflow.

The trade-off is control over quality versus performance. A coarser cage is cheaper and smaller, while a finer cage can follow the original animation more closely. The method is best suited to connectivity-preserving animation, such as swaying vegetation, grass, crowds, distant characters, or animation level of detail. It is less appropriate for cases where topology changes, very small-scale motion, or sharp character deformations are the visual focus. Please note though that tetrahedron-cage driven animations can also be used in the context of partitioned top-level structures and can also coexist with cluster level acceleration structures.

What’s next for tetrahedral cages

Overall, the paper presents tetrahedral cages as a practical bridge between rich animated content and hardware-accelerated ray tracing. By animating a compact proxy and reusing static rest-pose geometry, it opens the door to scenes that would otherwise be too expensive to update every frame. Future work will focus on improving cage generation and animation quality, as well as integration with modern ray-tracing pipelines. Please note that we are hard at work to create DXR samples and a header-only C++ library that allows building high-quality tetrahedral cages for skinned objects, keyframe-animated objects, and static objects. If you’re working on animated geometry at scale, we encourage you to dive into the full paper below, and stay tuned to GPUOpen as this research continues to evolve.

Download the Ray Tracing Massive Amounts of Animated Geometry paper and see more AMD graphics research on our Research Publications page.
Watch a video of the High-Performance Graphics 2026 presentation of this paper.

Footnotes

Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied. GD-97.

DirectX is a trademark of the Microsoft group of companies.

Hacker Times