Microsoft has released the Shader Model 6.10 preview, included in the new AgilitySDK 1.720-preview build. The update adds a streamlined algebra matrix API focused on direct control of gpu-dedicated Artificial Intelligence engines. The new API from the class linalg::Matrix exposes matrix operations to the shader language, giving developers a unified way to target hardware used for matrix multiplication and accumulation.
Modern gaming gpus already include dedicated hardware for Artificial Intelligence workloads, but access methods differ by vendor. Microsoft is positioning Shader Model 6.10 as a common abstraction layer across Tensor cores from NVIDIA, XMX cores from Intel, and Artificial Intelligence accelerators in AMD gpus. That approach is intended to let neural rendering operations run across multiple gpus with a single programming effort, reducing the need for separate vendor-specific implementations.
Microsoft said it is seeing a significant increase in graphics features that use neural network-based rendering techniques to improve image quality. That shift is expected to increase demand for matrix units in modern gaming gpus. By exposing all known matrix operations for popular gaming gpus from AMD, Intel, and NVIDIA, Shader Model 6.10 is designed to support broader adoption of these rendering techniques through DirectX 12.
Support is uneven across hardware vendors in the preview. The feature is supported across all NVIDIA RTX hardware, as it includes Tensor cores. Intel support is planned for an upcoming release, with B-series gpus expected to be compatible. On AMD hardware, only RDNA 4-based Radeon RX 9000 series gpus support this feature, with no support planned for older models like the RX 7000 series and below.