After the presentation of Nvidia GeForce RTX 5000 series video cards, Microsoft confirmed that it will add Neural Rendering functions to the DirectX API. It covers a wide range of new imaging techniques that apply artificial intelligence and machine learning to traditional graphics processing methods.
Image source: nvidia.com
Support for neural rendering in Microsoft DirectX will begin with the Cooperative Vectors solution – it will reveal the capabilities of neural shaders on Nvidia GeForce RTX 50 video cards. These are the manufacturer’s first video cards equipped with tensor cores with support for neural shaders.
«Cooperative Vectors accelerate AI workloads for real-time rendering. They allow you to multiply matrices by vectors of arbitrary size, optimizing matrix-vector operations. Cooperative Vectors also allow AI tasks to be executed at different stages of shader processing, allowing small neural networks to perform pixel shader operations without consuming the entire GPU. This allows you to implement neural graphics methods without stopping the graphics pipeline every time an AI/ML operation is performed,” Microsoft explained. In other words, AI is built into the graphics pipeline rather than treated as a standalone component that is calculated separately, allowing developers to flexibly use AI to improve specific aspects of rendering.
It is noted that neural shaders are currently only supported on Nvidia GeForce RTX 50 video cards, but Microsoft is also collaborating with AMD, Intel, and Qualcomm to provide support for new features on equipment from different suppliers – the software giant intends to implement them in the DirectX core.