DirectX Raytracing update standardizes multi-vendor GPU features
Microsoft's latest DirectX Raytracing update introduces Clustered Geometry and Partitioned TLAS, features that can cut ray tracing memory use by 90% and dramatically boost frame rates on supported hardware from AMD, Intel, and Nvidia.

Microsoft has detailed the biggest change to DirectX Raytracing since the API arrived in 2018. The update unifies Clustered Geometry and Partitioned TLAS into a single, multi-vendor path for AMD, Intel, and Nvidia GPUs.
Currently, a GPU must rebuild an object's acceleration structure from scratch whenever a mesh deforms or streams in at a new level of detail. The new DirectX Raytracing features aim to solve this inefficiency. Clustered Geometry splits models into small, reusable clusters of up to 256 triangles that are built once. Partitioned TLAS breaks the game world into separate regions, allowing the system to update only the areas where objects have moved. A third feature, GPU-driven acceleration structure builds, lets the GPU execute these operations directly from its own memory, bypassing the CPU. These tools can be used together or separately.
Performance and memory gains
Microsoft's own testing demonstrates the potential scale of improvement. In a test sample running on a Radeon RX 9070, the frame rate leapt from 1.2 fps to 79 fps. The time needed to build acceleration structures each frame plummeted from 788 milliseconds to just over 3 milliseconds.
Memory savings were equally dramatic. The video memory required for ray tracing structures shrank from roughly 7.5GB to 774MB, a reduction of about 90 percent. Further microbenchmark testing on a Radeon RX 9070 XT showed a 7.83x speed-up in per-frame GPU ray-tracing and a 89.7% cut in acceleration structure memory use.
| Test Metric | Before | After | Improvement |
|---|---|---|---|
| Frame Rate (Radeon RX 9070) | 1.2 fps | 79 fps | ~66x higher |
| Per-Frame Build Time | 788 ms | ~3 ms | Massive reduction |
| VRAM for RT Structures | ~7.5 GB | 774 MB | ~90% less |
| GPU Ray-Tracing Speed (Microbenchmark) | Baseline | 7.83x faster | 7.83x speed-up |
Hardware support and availability
No new hardware is required to use these features. Support is confirmed across current-generation GPUs from all major vendors.
| Vendor | Supported Hardware |
|---|---|
| Nvidia | Every RTX card (GeForce RTX 3060 and newer) |
| AMD | Radeon RX 9000 series |
| Intel | Arc B-series GPUs and recent Core Ultra chips |
| Qualcomm | Graphics solutions (day-one support) |
The features are currently in preview. Enabling them requires switching on Windows Developer Mode. Readers can expect day-one support in upcoming titles for the listed hardware.
Context and caveats
Microsoft warns that real games should not expect performance wins anywhere near the magnitude of its controlled test. The company stated in all caps that games shouldn't expect wins anywhere near that size. The test was designed to illustrate the maximum potential in a best-case scenario, but real-world gains will be tempered by increased needs for temporary scratch memory during processing.
In practice, commentators note that 'up to' values typically represent optimal lab conditions. For real games, users can generally expect a more modest improvement of a few percentage points in frame rate. The core benefit remains the massive reduction in memory overhead, which could enable more complex ray-traced effects within existing VRAM budgets.
These features will appear in supported titles across AMD Radeon RX 9000 series, Nvidia GeForce RTX 3060 and newer, and Intel Arc B-series GPUs.





