Microsoft DirectX 12 is getting one of its biggest ray tracing upgrades in years, and it could make a major difference for future PC games. With the new Agility SDK 1.722 preview, Microsoft has introduced major DirectX Raytracing improvements designed to reduce VRAM usage, speed up acceleration structure updates, and make complex real-time ray-traced scenes easier to handle on modern GPUs.
The update adds three important features: Clustered Geometry, Partitioned TLAS, and a GPU-driven API for building acceleration structures. Together, these technologies aim to solve one of the biggest problems in real-time ray tracing: the high cost of updating detailed, constantly changing scenes.
Ray tracing relies on acceleration structures to calculate how light behaves in a 3D environment. When a game world changes because of animation, destruction, streaming assets, or procedurally generated geometry, those structures often need to be rebuilt or updated. That process can consume a large amount of GPU time and memory, especially in scenes with many moving objects.
Clustered Geometry tackles this by breaking large geometry data into smaller reusable pieces. Each cluster can contain up to 256 triangles and 256 vertices. Instead of rebuilding an entire bottom-level acceleration structure from scratch, developers can reuse and recombine these smaller clusters. This can dramatically reduce redundant work in scenes with repeated or frequently changing geometry.
Partitioned TLAS focuses on the top-level acceleration structure, which represents the broader scene. Rather than updating the entire structure when only part of the world changes, PTLAS allows the GPU to update only the affected regions. In theory, this means performance costs scale with the amount of change in the scene instead of the total size of the environment.
Microsoft’s early test results show how powerful this approach can be in the right scenario. In a Clustered Geometry sample running on an AMD Radeon RX 9070, performance increased from just 1.2 frames per second to 79 frames per second compared with a traditional bottom-level rebuild method. GPU build time per frame dropped from 788 milliseconds to only 3.11 milliseconds.
The memory savings were also significant. Total acceleration structure memory, including the top-level structure, fell from 7,519 MiB to 774 MiB. That represents an 89.7% reduction in memory usage, which is especially notable as modern games continue to push higher-resolution assets, larger worlds, and more advanced lighting.
Another Microsoft demo involving a domino-style scene showed how Partitioned TLAS can reduce update time. The top-level update dropped from 2.88 milliseconds to 0.79 milliseconds per frame. However, the company also noted that ray traversal became slightly slower in that example, showing that these new techniques are not a universal performance boost in every situation.
While the numbers are impressive, players should not expect every game to suddenly see massive ray tracing gains. Microsoft has made it clear that these tests were designed as technology demonstrations, not real-world gaming benchmarks. Actual results will depend heavily on how a game engine is built, how much of the scene changes per frame, and how developers use the new tools.
There are also trade-offs. Clustered Geometry can require much more scratch memory during the build process. In Microsoft’s sample, static build scratch memory increased from 1.83 MiB to 526 MiB. The technique also does not support refits, meaning developers will need to carefully decide when and where it makes sense to use it.
The good news is that these DirectX Raytracing upgrades do not require brand-new ray tracing hardware. They are designed to work with existing GPUs that already support ray tracing, as long as the necessary software and driver support are available.
For now, the features are aimed at developers rather than everyday gamers. To experiment with them, developers need the DirectX 12 Agility SDK 1.722 preview runtime, the Shader Model 6.10 preview shader compiler, a compatible driver, and Windows Developer Mode enabled.
Current supported hardware includes Nvidia RTX graphics cards, AMD Radeon RX 9000 series GPUs, Intel Arc B-series graphics, and Intel Core Ultra 200V and Series 3 processors. Qualcomm is also working on driver support for Snapdragon X2, with availability targeted for the first half of 2027. Support for some older ray tracing GPUs has not yet been confirmed.
Microsoft has not announced when these features will become available in a mainstream retail release. Even so, this update marks an important step for the future of DirectX 12, PC gaming graphics, and real-time ray tracing performance.
If widely adopted by game developers, Clustered Geometry and Partitioned TLAS could help future games deliver richer ray-traced worlds with lower VRAM requirements and faster scene updates. That could be especially important as game environments become larger, more dynamic, and more visually demanding.
