Microsoft’s new DirectX Raytracing upgrade shows huge gains on AMD Radeon RX 9070 XT
Microsoft’s latest DirectX Raytracing update is already showing impressive potential, especially for future games with extremely dense geometry and advanced ray tracing. Early independent testing on an AMD Radeon RX 9070 XT suggests that the new DXR features can dramatically improve acceleration structure performance while cutting memory usage by a massive amount.
The upgrade arrives through Microsoft’s DirectX 12 Agility SDK 1.722 Preview, which introduces several important ray tracing technologies: Clustered Geometry, Partitioned Top-Level Acceleration Structures, and GPU-driven acceleration structure builds. Together, these features are designed to make ray tracing more efficient in scenes packed with detailed objects, high triangle counts, and complex environments.
After AMD confirmed support for these new DirectX Raytracing features on RDNA 4 graphics cards, graphics enthusiast Osvaldo Pinali Doederlein tested Microsoft’s claims using an AMD Radeon RX 9070 XT. The results were striking.
In testing with Microsoft’s D3D12RaytracingClusteredGeometry sample and AMD’s Agility SDK Developer Preview driver, Osvaldo compared the traditional method of rebuilding Bottom-Level Acceleration Structures with the newer Clustered Geometry approach. The test used a fixed camera position at 1600 × 920 resolution and a geometry-heavy workload designed to resemble Microsoft’s original demonstration.
The traditional BLAS method delivered only 1.7 FPS. With Clustered Geometry enabled, performance jumped to 91 FPS.
That is roughly a 54x increase in frame rate.
Memory usage also improved significantly. Acceleration structure memory dropped from 7,518 MB with the traditional method to just 771 MB using Clustered Geometry. That represents nearly a 90% reduction in memory consumption.
The reported rebuild time showed an even more extreme difference, falling from around 600 milliseconds to just 2.6 microseconds. Ray dispatch time, however, changed only slightly, moving from 8.5 milliseconds to 8.2 milliseconds.
This detail is important. The major performance improvement does not come from making ray traversal itself dramatically faster. Instead, the big gain comes from reducing the heavy cost of rebuilding ray tracing acceleration structures in complex scenes.
In ray tracing, the Top-Level Acceleration Structure organizes scene instances, while the Bottom-Level Acceleration Structure stores the underlying geometry. When scenes become extremely detailed, rebuilding these structures can become expensive in both performance and memory. Microsoft’s Clustered Geometry technology is designed to reduce that bottleneck.
However, these results should not be mistaken for a guarantee that upcoming games will suddenly run path tracing at extremely high frame rates on an RX 9070 XT.
Osvaldo also tested more realistic and less extreme configurations. With the sample’s default setting of 100 additional BLASes, Clustered Geometry was about twice as fast. The massive 50x-plus improvement appeared only after increasing the workload to 10,000 additional BLASes, effectively multiplying the geometry load by 100.
Resolution also affects the performance uplift. At 4K, the same geometry-heavy configuration improved from 1.6 FPS to 28 FPS. That is still a major 17.5x improvement, but it is much smaller than the result at the lower test resolution.
There were also cases where ray dispatch performance appeared to regress by as much as 40%. Even with that drawback, Clustered Geometry still remained faster overall in the tested scenarios.
The test scene reportedly contained around 91 million triangles. For comparison, highly advanced modern games can push geometry far beyond what older engines were designed to handle. Remedy Entertainment previously revealed that Alan Wake II can feature around 200 million triangles before culling in its most complex scenes, spread across millions of meshlets and hundreds of thousands of instances. While the workloads are not directly comparable, it shows why more efficient geometry handling is becoming increasingly important for real-time ray tracing.
The biggest takeaway is that Microsoft’s DXR improvements could help future PC games handle much denser worlds without overwhelming GPU memory or spending huge amounts of time rebuilding acceleration structures. This could be especially valuable as developers continue to increase scene complexity, object density, and ray-traced visual effects.
Still, real game performance gains are expected to be more modest than the synthetic benchmark results suggest. Games include many other rendering tasks, such as shading, lighting, post-processing, animation, physics, streaming, and CPU-side work. Developers also need to integrate these new features at the engine level, which can require significant effort.
There is also the question of hardware support. AMD’s current preview driver supports RDNA 4 GPUs, including the Radeon RX 9000 Series, through driver version 26.20.12.01. Osvaldo suggested that AMD’s implementation may rely on RDNA 4 architectural improvements such as BVH8 and Oriented Bounding Boxes, which could make support for older Radeon graphics cards less likely.
Microsoft says the standardized APIs do not fundamentally require new ray tracing hardware, but actual support will depend on GPU vendors, drivers, and implementation choices.
For AMD Radeon users, the early signs are encouraging. The Radeon RX 9070 XT test shows that Microsoft’s new DirectX Raytracing technologies can deliver huge improvements in the right workloads, especially when geometry complexity becomes the main bottleneck.
The real test will come when game developers begin using Clustered Geometry and related DXR features in full production engines. If these tools translate well into real games, future ray-traced and path-traced titles could become more efficient, less memory-hungry, and better suited for increasingly detailed virtual worlds.
