 |
| Project PICA PICA from SEED. |
Microsoft has unveiled what it's calling the most significant update to DirectX Raytracing (DXR) since the technology debuted in 2018, and the numbers are genuinely staggering. The Agility SDK 1.722 preview, released on October 28, 2026, introduces Clustered Geometry and Partitioned Top-Level Acceleration Structures (PTLAS)—two features that could fundamentally change how games handle ray tracing in complex, dynamic scenes.
According to Microsoft's official DirectX Developer Blog, these new capabilities aim to dramatically reduce the cost of ray tracing detailed, changing environments. And based on early testing, the results speak for themselves.
The VRAM Problem Solved?
One of the biggest barriers to widespread ray tracing adoption has been its insatiable appetite for VRAM. Traditional bottom-level acceleration structures (BLAS) need to be rebuilt whenever geometry changes—whether that's due to animation, streaming, or level-of-detail transitions. For large, dynamic worlds, this creates a perfect storm of memory consumption and performance bottlenecks.
Microsoft's solution breaks this problem into smaller, more manageable pieces. Clustered Geometry divides geometry into small, reusable clusters of up to 256 triangles and 256 vertices, which can be recombined rather than rebuilt from scratch. Meanwhile, Partitioned TLAS tackles scene-wide updates by focusing only on regions that have changed, keeping the rest of the acceleration structure intact.
The Numbers That Matter
In a test conducted on an AMD Radeon RX 9070, Microsoft demonstrated the potential of these technologies. The results, while achieved in a controlled test environment, are eye-opening:
| Metric | Traditional BLAS | Clustered Geometry | Improvement |
|---|---|---|---|
| Frame Rate | 1.2 FPS | 79 FPS | ~66x higher |
| GPU Build Time | 788 ms | 3.11 ms | ~254x faster |
| Total AS Memory | 7,519 MiB | 774 MiB | 89.7% reduction |
The memory savings alone are remarkable—nearly 90% less VRAM required for acceleration structures. For gamers with 8GB or 12GB graphics cards, this could mean the difference between playable ray-traced experiences and slideshows.
A separate domino simulation test showed PTLAS cutting top-level update times from 2.88 ms to 0.79 ms per frame, though ray traversal performance took a slight hit—a trade-off Microsoft acknowledges may be worthwhile in many scenarios.
 |
| DirectX Raytracing performance. |
What This Means for Real Games
Before you get too excited, Microsoft is quick to temper expectations. The company explicitly warns developers not to assume games will see gains of this magnitude, calling the test scene a "test playground." There are also trade-offs: Clustered Geometry requires significantly more scratch memory during builds (static build scratch jumped from 1.83 MiB to 526 MiB), and the technique doesn't support refits.
According to the DirectX Raytracing Functional Spec, these features work on existing ray tracing hardware with a driver update—no new GPU required. However, the spec notes that newer hardware may show additional improvements, and support for older ray tracing GPUs is still "to be determined."
Hardware Support and Availability
The features are currently in preview, requiring Windows Developer Mode, the Agility SDK 1.722 preview runtime, and a supporting driver. Current hardware support includes:
- NVIDIA: All RTX GPUs
- AMD: Radeon RX 9000 series
- Intel: Arc B-Series and Core Ultra 200V/Series 3 processors
- Qualcomm: Snapdragon X2 support targeted for Q1/Q2 2027
Notably, Gears of War: E-Day is already adopting clustered geometry through NVIDIA's RTX Mega Geometry API, suggesting the technology is moving from theory to practice.
The Bigger Picture
This update represents a philosophical shift in how DirectX handles ray tracing. As Microsoft's 2018 DXR announcement noted, the original vision was to "bridge the gap between the rasterization techniques employed by games today, and the full 3D effects of tomorrow." Eight years later, that tomorrow is arriving—and it's arriving with dramatically lower memory requirements.
For developers, this means the ability to ray trace full-detail geometry rather than simplified proxies, leading to more accurate shadows and reflections. For gamers, it could eventually mean ray tracing becomes viable on more modest hardware, without requiring a $1,500 GPU.
The preview is available now for developers to experiment with, and Microsoft is actively soliciting feedback before the final release. For those interested in diving into the technical details, the full specification and developer blog post provide comprehensive documentation.
No retail release date has been announced, but with Shader Model 6.10 expected to hit retail in early December, the pieces are falling into place for a significant leap forward in real-time ray tracing.
Sources & Further Reading:
- Microsoft DirectX 12 update slashes ray tracing VRAM usage by 90%
- [PREVIEW] Raytracing: Clustered Geometry, Partitioned TLAS, Indirect Acceleration Structure Builds
- Announcing Agility SDK 1.620 Retail and 1.722 Preview
- DirectX Raytracing (DXR) Functional Spec, Part 2
- Announcing Microsoft DirectX Raytracing!