Microsoft has released a preview of DirectX Raytracing introducing Clustered Geometry and Partitioned Top Level Acceleration Structures. The architecture breaks 3D meshes into clusters of up to 256 triangles and isolates top level acceleration updates to changed regions of a scene. The software is available now through the Agility SDK preview with driver support across current graphics silicon from AMD, Intel, and NVIDIA.
Microsoft detailed the technical preview on the official DirectX Developer Blog, marking a major architectural shift toward GPU execution. Standard raytracing requires complete rebuilds of large bottom level acceleration structures whenever geometry deforms or streams. Clustered Geometry divides meshes into Cluster Level Acceleration Structures containing up to 256 vertices and 256 triangles. This approach allows game engines to raytrace the exact full detail models used during rasterization instead of low quality proxy geometry.
Partitioned TLAS manages the broad scene hierarchy. Rather than rebuilding the entire top level tree when a few objects move, the system organizes entities into distinct spatial partitions. Stationary regions carry over untouched between frames, while moving objects update locally. A single graphics card execution call manages the entire operation buffer through indirect arguments, removing workload overhead from the central processor.
Animated characters rely on cluster templates that define topology in advance, allowing compute shaders to calculate new vertex coordinates per frame. The entire toolchain functions through Shader Model 6.10 within the experimental Agility SDK runtime. While studios previously tested proprietary implementations in titles like Gears of War E Day, this release establishes a single unified standard across the industry.
Hardware compatibility applies to all NVIDIA RTX cards, AMD Radeon RX 9000 graphics processors, and Intel Arc B Series silicon alongside Core Ultra Series 3 processors. Qualcomm plans to roll out driver support for Snapdragon X2 devices later. In AMD test sandbox benchmarks running on an RX 9070, clustered geometry increased frame rates from 1.2 to 79 frames per second while reducing animated acceleration memory use by 96.4%.
