GDC 2026: Evolving DirectX for the ML Era on Windows

Machine learning is becoming foundation to real-time graphics, and DirectX is evolving to meet that shift across shader-level and model-level ML. These systems are increasingly embedded throughout the graphics pipeline, influencing how developers generate frames, author content, and realize artistic visions. At GDC 2026, the DirectX team shared more about this evolution during the Xbox Dev Summit session titled “Evolving DirectX for the ML Era on Windows.” We’ve got some highlights below, but be sure to head over to the DirectX Developer Blog for their full announcement.

Introducing DX Linear Algebra

Last year, DirectX took a major step into the ML era with the introduction of Cooperative Vector in Shader Model 6.9. For the first time, developers could access hardware-accelerated vector-matrix operations directly from HLSL, enabling a class of neural rendering techniques that execute inline with traditional shading. These workloads, such as neural texture compression (DTC) and neural radiance caching (NRC), map naturally to highly parallel, per-pixel inference.

After analyzing expanded ML usage, we found not all workloads fit Cooperative Vector’s execution model. Super resolution, denoising, temporal upscaling, and general LLM inference require matrix-matrix operations, shared data across threads, and batch-oriented execution. Vector-matrix primitives alone cannot always efficiently express these.

To address this gap, we’re introducing DirectX Linear Algebra, designed to support both vector and matrix-based ML workloads under a single programming model. DX Linear Algebra adds first-class matrix-matrix operations while preserving the ability to author ML directly in HLSL, giving developers explicit control over math, data flow, and execution for shader-level ML scenarios.

Expanding to Model Level ML with DirectX Compute Graph Compiler

While shader-level ML is powerful, many modern ML-driven graphics workloads are best expressed and optimized as full computation graphs, not as isolated operators or hand-authored kernels. These graphs capture end-to-end structure—dataflow, dependencies, and deep fusion—that are difficult to exploit at the shader level.

We have introduced DirectX Compute Graph Compiler, a new DirectX ML compiler API that executes full-model graphs with native-class GPU performance.

Models flow from modern frameworks, where DirectX can analyze and specialize the complete graph for a given device before lowering it into optimized workloads that integrate natively with D3D12 queues and command lists.

Key benefits include:

• Dropping full models into engines without shader rewrites
• Automatic graph optimization, memory planning, and operator fusion
• Portable performance across IHVs
• Unified tooling with PIX, showing graphics and ML workloads in a single capture

Shader-level ML and model-level ML now live side-by-side in DirectX, including HLSL Linear Algebra for small, inline workloads and DirectX Compute Graph Compiler for larger models.

Support from hardware vendor partners

DirectX partners AMD, Intel, NVIDIA, and Qualcomm have all committed to support features like DirectX Linear Algebra and DirectX Compute Graph Compiler and have shared how these partnerships will benefit developers and push the industry forward. To read their full quotes, visit the DirectX Developer Blog.

What’s Next

DirectX Compute Graph Compiler will be available for private preview this summer; to join, please reach out to your Windows representative.

DX Linear Algebra will enter public preview in April, giving developers an early opportunity to help shape the future of ML‑assisted graphics. See the Linear Algebra spec for more detail.

We’re excited to continue this ML graphics generation journey with our partners and the developer community. Check out our GDC session at aka.ms/GDC2026ML, and stay tuned to the DirectX blog for deeper dives, samples, and updates.