Microsoft Expands DirectX 12 With Shader Model 6.10 and AgilitySDK 1.720

Microsoft is pushing DirectX 12 forward with new preview releases that target performance, AI workloads, and developer flexibility, as Wccftech reports. The latest Shader Model 6.10 and AgilitySDK 1.720 introduce deeper GPU control, modern rendering tools, and improved parallel execution.

The update builds on earlier releases like Shader Model 6.9 and DXR 1.2, signaling a continued focus on real-time graphics and machine learning integration.

Linear algebra APIs bring AI into shaders

Shader Model 6.10 introduces LinAlg, a new set of linear algebra APIs designed for GPU shaders. These APIs allow developers to run matrix operations directly inside shaders, which opens the door for real-time neural rendering.

This change improves performance for machine learning tasks, image processing, and advanced visual effects. Higher bandwidth matrix multiply-accumulate operations also help accelerate complex workloads without relying on external compute pipelines.

Better control with new wave-level intrinsics

Microsoft added new wave-level intrinsics, including GetGroupWaveIndex() and GetGroupWaveCount(). These functions give developers visibility into how threads are grouped and executed on the GPU.

Developers can now optimize workloads more safely across different GPU architectures. The update removes the need for fragile workarounds that depended on fixed wave sizes, improving portability and stability.

Shared memory limits removed for complex workloads

Another key improvement targets group shared memory. Shader Model 6.10 removes the long-standing 32 KB limit, which previously restricted more advanced rendering techniques.

Developers can now query available memory at runtime using MaxGroupSharedMemoryPerGroup and define limits at compile time with the new GroupSharedLimit attribute. This enables more demanding scenarios like large-scale tile culling and heavy matrix computations.

DirectX 12 gains parallel execution improvements

Alongside shader updates, Microsoft introduced changes to Direct3D 12 through batched asynchronous command lists. This update removes the requirement for strictly sequential command execution.

Independent GPU tasks can now run in parallel, reducing stalls and improving overall performance. Developers only need to handle synchronization when tasks depend on each other, which simplifies optimization.

Broad hardware support across GPU vendors

The new features already show wide industry support. NVIDIA enables most capabilities across RTX GPUs, while AMD supports them on RDNA 3 and RDNA 4 architectures. Intel also supports parts of the update on Arc B-Series GPUs, with additional features planned.

Some capabilities remain in progress. For example, NVIDIA will add support for certain wave-level features later, while Intel plans to expand compatibility for the new linear algebra APIs.

A step toward AI-driven graphics pipelines

This update reflects a broader shift in graphics development. Microsoft continues to merge traditional rendering with AI and machine learning, giving developers more tools to build advanced visual experiences.

By improving GPU efficiency, removing legacy limits, and enabling real-time AI workloads, Shader Model 6.10 and AgilitySDK 1.720 mark another step toward more flexible and powerful graphics pipelines.

In related developments, NVIDIA has been working on a DLSS 4.5 over-the-air update alongside a new Streamline SDK, while AMD recently released FSR 4.1 with its own SDK, highlighting growing competition in AI-powered rendering technologies.