DirectX 2026 Machine Learning Updates and System Architecture
At GDC 2026, Microsoft detailed fundamental changes to the DirectX architecture. Machine learning (ML) is moving from a post-processing utility to a core element of real-time graphics pipelines. This evolution is being driven by two key technologies: DX Linear Algebra and DirectX Compute Graph Compiler, aimed at supporting both shader-level and model-level ML workloads.
DX Linear Algebra enables hardware-based ML operations through its ability to create a direct path from shader operations to machine learning processing workloads. HLSL developers will now have access to matrix-matrix operations through DX linear algebra, which uses Shader Model 6.9 and cooperative vectors as a foundation. The new development provides a visual area for developers to perform ML procedures along with standard shading functionality.
Matrix functionality now allows users to process matrix-matrix operations through hardware acceleration beyond previously existing vector-matrix operations capabilities. Applications Advanced denoising, temporal upscaling, and neural radial caching represent pixel-wise inference tasks that this program can perform with optimal performance. This ability to manage data flow and math operations allows developers to coordinate data processing and math operations across all shaders.
DirectX Compute Graph Compiler allows processing of full machine learning models, and linear algebra processes shader functions at a reduced scale. A new application programming interface allows users to evaluate the complete computational graph while creating system-specific GPU profiles, improving system performance across all PC components.
Native integration allows game engines to import and use models directly, eliminating the need for shader rewriting tasks. The compiler completes memory design, operator merging, and graph optimization tasks through an automated function system. The integrated D3D12 interface provides performance that operates independently from hardware systems from AMD, Intel, NVIDIA, and Qualcomm vendors. Integrated debugging allows users to create a single system capture that displays both graphics and ML workloads with full PIX tool compatibility.
The DirectX update has received support from all major hardware manufacturers who have confirmed support for the update. AMD, Intel, and NVIDIA will begin supporting DX linear algebra on day one, while Qualcomm will help develop the Compute Graph Compiler, which makes model-level ML work like a native technology in real-time engines. GPU design architecture partnerships ensure that ML-based rendering remains operational reliably across all supported GPU design systems.
Deployment of these technology solutions will begin in the second quarter of 2026.
- DX Linear Algebra DX Linear Algebra public preview begins in April 2026
- DirectX Compute Graph Compiler DirectX Compute Graph Compiler will begin private preview access during Summer 2026.
This update creates a new environment where ML acts as the primary rendering component, allowing developers to combine traditional rasterization and neural rendering techniques using new creative tools.
