DirectX Updates: Linear Algebra, Graph Compiler, and DXR 2.0
Microsoft has released DirectX updates introducing DirectX Linear Algebra (DXLA) and DirectX Compute Graph Compiler. These tools simplify integrating machine learning algorithms into shaders for games and apps. DXLA adds linear algebra instructions at the shader level, including matrix multiplication—a key operation for upscaling, anti-aliasing, and real-time ray tracing. Compute Graph Compiler integrates with DirectML, enabling ML models to run as standard modules without manual GPU optimization.
DirectX Linear Algebra: Mathematical Primitives for Shaders
DXLA extends DirectX's math toolkit. Shaders now support built-in linear algebra operations:
- Matrix multiplication of various ranks.
- Vector-matrix operations.
- Determinants and inverses for small matrices.
This is crucial for ML tasks in graphics: neural networks for DLSS-like upscaling or denoising require intensive matrix computations. Developers can write shaders with native support for these operations, avoiding emulation through basic HLSL instructions.
Public preview of DXLA is coming in April 2026. Support from AMD, Intel, NVIDIA, and Qualcomm is confirmed.
DirectX Compute Graph Compiler: Automating ML on GPU
Compute Graph Compiler is a high-level interface for DirectML. It compiles compute graphs of ML models into optimized code for specific hardware. Key features:
- Automatic adaptation to GPU architectures (NVIDIA, AMD, Intel).
- Support for standard model formats without custom optimization.
- Integration into the graphics pipeline as a compute shader.
The system analyzes the graph, distributes operations across execution units, and minimizes memory traffic. This simplifies porting ML features to cross-platform apps. Preview expected in summer 2026.
Advanced Shader Delivery for PC
Advanced Shader Delivery (ASD) is moving from portable consoles to desktop. The technology automates shader pre-compilation, eliminating stuttering at game startup. The process:
- Pre-compilation on the server or locally.
- Delivery of ready binaries.
- Minimal runtime compilation.
NVIDIA will roll out ASD for GeForce RTX by the end of 2026. AMD and Intel have announced similar support. It's especially relevant for complex pipelines with ray tracing.
DirectX Raytracing 2.0 and Opacity Micromaps
DXR 2.0 requires Shader Model 6.10 and introduces Opacity Micromaps (OMM) for optimizing semi-transparent objects. OMM speeds up BVH traversal for foliage, hair, and particles:
- Opacity micromaps instead of full geometry checks.
- 2–4x reduction in RT-core load according to Microsoft tests.
- Compatibility with DXR 1.x.
The update integrates with DXLA for ML-based ray denoising.
What Matters
- DXLA adds matrix operations to shaders, simplifying ML in graphics.
- Compute Graph Compiler automates DirectML optimization for GPUs.
- ASD eliminates initial stuttering on PC with support from all vendors.
- DXR 2.0 with OMM optimizes semi-transparent scenes for real time.
- Timeline: DXLA in April, Compiler in summer, ASD in the second half of 2026.
— Editorial Team
No comments yet.