Microsoft used GDC 2026 to tease a major leap forward for Windows gaming, unveiling ML-powered DirectX features and a new approach to shader delivery designed to cut down on stutter and painfully long load times. On top of that, the company also outlined early details for DXR 2.0, the next evolution of DirectX Raytracing, which is expected to arrive with Shader Model 6.10 support.
DirectX is going all-in on machine learning
Machine learning is increasingly becoming the backbone of modern rendering. From denoising and temporal upscaling to advanced reconstruction techniques, today’s graphics pipelines are leaning harder on matrix-heavy workloads that don’t always fit neatly into traditional shader execution models. Microsoft’s answer is to expand DirectX so developers can integrate ML more naturally and efficiently.
A key step is Cooperative Vectors in Shader Model 6.9, already available through the Agility SDK 1.619 release. This opens the door to neural rendering techniques such as Neural Texture Compression and Neural Radiance Caching. Support is already in place on NVIDIA’s Blackwell architecture, while AMD has indicated it’s working toward support in future GPU architectures.
To better handle the growing demand for matrix and vector math in real-time graphics ML workloads, Microsoft is also introducing DirectX Linear Algebra. The goal is a single programming model that supports both vector-based and matrix-based ML workflows, giving developers explicit control over the math, data flow, and execution behavior for shader-level ML scenarios. In short, it’s meant to make GPU-accelerated ML inside graphics pipelines more practical, more controllable, and ultimately faster.
Microsoft is also adding a DirectX Compute Graph Compiler, described as a new DirectX ML compiler API that can execute full model graphs with native-class GPU performance. The idea here is to reduce the friction of using larger ML models in games and engines without forcing extensive shader refactoring. Some of the highlighted advantages include unified profiling and debugging with PIX (so graphics and ML workloads can be inspected together), the ability to drop full ML models into engines without shader rewrites, and automated optimization features such as memory planning and operator fusion. Microsoft is also emphasizing portability across different GPU vendors.
The company says the Compute Graph Compiler is expected to be available as a private preview this summer, while DirectX Linear Algebra is planned for public preview in April.
Advanced Shader Delivery aims to eliminate shader stutter and reduce load times
Alongside the DirectX ML announcements, Microsoft also detailed Advanced Shader Delivery for Windows, a feature already seen in action on Xbox ROG Ally and Ally X devices. The broader PC gaming goal is straightforward: stop making players suffer through shader compilation stutters and long “first run” waits.
The concept is to let gamers download fully compiled shaders in advance for their specific PC hardware, rather than relying on compilation during gameplay or at launch. Microsoft describes this as an ecosystem effort spanning game developers, GPU vendors, and game storefronts.
Agility SDK 1.619 adds two APIs designed to support this system:
The App Identity API, which allows applications to declare a standardized application identity to D3D12 and graphics drivers before a D3D12 device is created. This identity also becomes important for associating shader database assets properly, and Microsoft notes that attaching application identity to the SODB will be required when submitting an SODB file through the Xbox Partner Center.
The Stats API, which gives developers insight into how well a precompiled shader database (PSDB) performs, including metrics such as shader cache hit rate for specific hardware configurations. This is meant to help teams validate whether their precompiled shader strategy is working in the real world across different PCs.
To enable support, developers will need to integrate SODB collection in the game engine and submit it with the game package through the Xbox Partner Center. Hardware vendors are already signaling support across platforms, and Microsoft says NVIDIA plans to bring Advanced Shader Delivery to GeForce RTX users later this year, with AMD and Intel also working to expand support more broadly.
DXR 2.0 is in the works with Shader Model 6.10
Microsoft also confirmed it’s developing DXR 2.0, the next major version of DirectX Raytracing. According to the early outline, a device will be considered DXR 2.0 compliant if it supports Opacity Micromaps along with Shader Model 6.10. Microsoft also described partial support paths for hardware that may lack Opacity Micromaps but retains other capabilities, with compliance details tied to updated tiering and feature caps.
Shader Model 6.10 requirements include support for TriangleObjectPositions, and Microsoft notes that some hardware may be able to gain certain DXR 2.0-related capabilities via driver updates, though not necessarily all features required for higher tiers.
These DXR 2.0 features are still under development, with Microsoft suggesting a preview timeframe around late summer 2026. The specs are being shared early because the feature direction is being aligned across hardware vendors, though Microsoft cautions the details may still be refined before release.
Big picture: what this means for PC gamers and developers
Taken together, Microsoft’s GDC 2026 announcements point to two major priorities for the next wave of Windows gaming: making ML-driven rendering easier and faster to implement through DirectX, and improving real-world playability by tackling shader compilation stutter and load-time frustrations at the ecosystem level. If these rollouts land as planned, the payoff could be smoother launches, fewer hitching issues, and more advanced visual techniques becoming mainstream across a wider range of PC hardware.
