마이크로소프트, 2018년 이후 최대 DirectX 레이트레이싱 업그레이드 공개, 엔비디아·AMD·인텔 GPU에 효율적 레이트레이싱 제공

Microsoft has announced the release of DirectX 12 Agility SDK 1.620 Retail and 1.722 Preview, introducing significant improvements to ray tracing, GPU efficiency, and Advanced Shader Delivery. The company describes the latest DirectX Raytracing (DXR) enhancements as its most substantial update since the technology debuted in 2018. New Clustered Geometry And Partitioned Acceleration Structures Could Significantly Improve Ray Tracing Efficiency The headline additions in Agility SDK 1.722 Preview are Clustered Geometry, Partitioned Top-Level Acceleration Structures (PTLAS), and GPU-driven acceleration structure builds, all designed to reduce the computational overhead associated with ray tracing complex, dynamic environments. Traditionally, ray tracing requires constructing and updating acceleration structures — also known as ray tracing bounding volume hierarchies (BVHs) — that represent scene geometry. This process can become particularly expensive in highly detailed environments featuring extensive animation, geometry streaming, and frequent changes to objects. Microsoft's new Clustered Geometry implementation addresses this by splitting geometry into smaller, reusable clusters containing up to 256 triangles and vertices each. These clusters can be assembled into larger acceleration structures without repeatedly rebuilding geometry from scratch. This could allow developers to use the same highly detailed geometry for both rasterization and ray tracing, potentially improving the accuracy of reflections and shadows without excessive acceleration structure construction costs. Meanwhile, PTLAS allows developers to update only the portions of a scene that have changed, rather than rebuilding its entire top-level acceleration structure. Combined with GPU-driven builds, this should improve efficiency in large, dynamic game worlds. Microsoft also demonstrated the potential performance improvements of Clustered Geometry using its D3D12RaytracingClusteredGeometry sample, which allows developers to compare traditional Bottom-Level Acceleration Structure (BLAS) rebuilding against Clustered Geometry in identical scenes. In one configuration running on an AMD Radeon RX 9070, switching to Clustered Geometry increased performance from just 1.2 FPS to 79 FPS (roughly 66x faster), while reducing acceleration structure build times from 788 ms to 3.11 ms. Total acceleration structure memory consumption also dropped by nearly 90%, from approximately 7.3 GiB to just 774 MiB, although scratch memory requirements increased substantially. Microsoft emphasizes that these results come from a specific development sample and configuration, so performance improvements in actual games will depend heavily on the workload, implementation, and GPU architecture. Interestingly, Microsoft highlights NVIDIA's RTX Mega Geometry implementation in Gears of War: E-Day as an early example of similar technology. The new standardized DirectX implementation could make such techniques more widely accessible across GPU vendors. The preview currently supports: - NVIDIA: All GeForce RTX GPUs. - AMD: Radeon RX 9000 Series. - Intel: Arc B-Series and supported Core Ultra integrated GPUs. Microsoft confirms that these features require no fundamentally new hardware capabilities beyond existing ray-tracing support, although driver support remains necessary. Agility SDK 1.620 Also Improves Advanced Shader Delivery Beyond ray tracing, the retail Agility SDK 1.620 introduces several improvements to Microsoft's Advanced Shader Delivery technology, which we've previously covered extensively. Notably, Partial Graphics Programs allow developers to reuse precompiled shader components across multiple pipeline configurations, potentially reducing shader package sizes and compilation overhead. Advanced Shader Delivery is now broadly supported across the Windows 11 ecosystem, including compatible GPUs from AMD, Intel, NVIDIA, and Qualcomm, although the new Partial Graphics Programs optimization currently remains exclusive to Radeon RX 7000 and RX 9000 GPUs, with support from other vendors planned. Additional improvements include DirectX Dump Files for diagnosing GPU crashes, Fence Barriers Tier 1, Batched Asynchronous Command List APIs, and new DirectX Compute Graph Compiler interfaces for machine-learning workloads. Finally, Microsoft confirmed that Shader Model 6.10 remains in preview, with its retail release planned alongside Agility SDK 1.621 in early December. While these developments won't automatically improve performance in existing games, they provide developers with more efficient tools for implementing increasingly sophisticated rendering technologies in future PC titles. News Sources: Microsoft DirectX Dev Blogs (1), Microsoft DirectX Dev Blogs (2) Follow Wccftech on Google to get more of our news coverage in your feeds.