Microsoft’s Agility SDK: Raytracing boost through SER & OMM – NVIDIA hardware benefits in particular

With Agility SDK 1.717, Microsoft has not just implemented a few new functions – it has poured a foundation on which the next generation of real-time graphics will be built. In an era in which ray tracing is mutating from a prestige feature to a baseline technology, Redmond is focusing on efficiency, parallelism and intelligence in the rendering process. With Shader Execution Reordering (SER) and Opacity Micromaps (OMM), two features are flowing into the DirectX pipeline that not only sound interesting in theory, but also manifest themselves measurably in FPS and flows – especially on NVIDIA’s RTX 50 hardware.

Shader execution reordering: parallelism without chaos

Traditionally, shader calls are executed in the order in which they occur. This sounds democratic, but is a performance disaster for highly complex, divergent workloads such as path tracing. SER breaks through this rigidity: developers can specifically tell the GPU which threads should be sorted and how – in order to utilize execution units more evenly and avoid divergent paths. The result: significantly increased coherence and up to twice the performance in path-driven engines. What used to be an incoherent chicken coop now becomes a well-ordered parade. The GPU works more efficiently, not harder. According to Microsoft, NVIDIA GPUs in particular, with their deeply staggered pipelines and specialized control logic, are predestined to benefit from SER. Initial benchmarks show a performance increase of around 80 % in Cyberpunk 2077 RTX Overdrive, and up to 2x more FPS are possible in Portal RTX – a brute performance boost that shows how much unused potential has been wasted in the shader queues so far.

Opacity Micromaps: Transparency without overhead

The second major feature, Opacity Micromaps, tackles an old problem in ray tracing: dealing with alpha transparencies. Until now, every potentially transparent pixel had to be laboriously checked by the so-called AnyHit shader – an expensive, often superfluous effort. OMMs avoid this bottleneck by using micromap data to tell the GPU in advance which areas are actually opaque or transparent. The hardware can then intelligently decide whether an AnyHit call is necessary at all. This is not only efficient, but also elegant – and saves enormous resources in complex scenes with many alpha-tested assets, such as foliage, window glass or fog. Microsoft claims performance gains of up to 2.3x – and NVIDIA is already technically in pole position here too. The RTX-50 series already has the appropriate hardware path, while AMD and Intel are still working on full implementation.

Cooperative Vectors: AI moves into the shader

Apart from ray tracing, the Agility SDK also brings a rather subtle but strategically important innovation: Cooperative Vectors. This function allows matrix and vector operations to be executed in a hardware-accelerated and synchronized manner between shaders – a boon for AI-based effects such as neural upscaling, temporal composition or physics-based image synthesis. This opens the door for Microsoft to integrate neural rendering directly into classic render pipelines – without having to rely on external engines or APIs. NVIDIA’s Tensor Cores in particular, which have been massively upgraded in the RTX 50 generation, benefit directly from this. Developers can now incorporate specific inferencing models into shader stages without getting tangled up in performance or timing problems. This is nothing less than preparation for a future in which classic rasterization and neural rendering merge.

Microsoft delivers – NVIDIA triumphs

The Agility SDK 1.717 is a technical game changer. Microsoft is not only delivering feature updates, but is also building on the strengths of modern hardware – while leaving the door open for future developments. While AMD and Intel are still working on full integration, NVIDIA GPUs are already ready – and are already delivering impressive results. Whether in AAA games, industrial visualizations or complex render farms: SER and OMM mark a paradigm shift in graphics processing. The GPU is no longer just getting faster – it is getting smarter. And that is the real progress.

Source: Microsoft