NVIDIA RTX Kit Finally Released, Features New “RTXNS” Neural Shading & More

After a short delay beyond its anticipated late January release, NVIDIA has officially rolled out the NVIDIA RTX Kit to the public, making most components available for developers. All libraries in the suite can now be accessed via GitHub, each with its own dedicated repository. Among these cutting-edge tools is RTX Neural Shaders (RTXNS), which aims to help developers integrate machine learning (ML) directly into graphics rendering pipelines.

A Quick Overview of RTX Kit Components

RTXNS uses the Slang shading language and leverages the Vulkan Cooperative Vectors extension, tapping into GPUs’ ML acceleration capabilities. This sample code demonstrates how to train neural networks for graphics tasks and deploy them in tandem with normal rendering steps.

Other prominent tech within the RTX Kit includes:

RTX Neural Texture Compression (NTC): Reduces memory usage by applying AI inference for texture compression. In a quick test by Compusemble, memory consumption was cut from 272MB down to 11.37MB, though it came with an associated performance cost.

RTX Mega Geometry (RTXMG): Already implemented in Alan Wake 2, this technology has a new demo showcasing how to accelerate complex geometry using a path-traced approach.

RTX Character Rendering (RTXCR): Provides examples of rendering path-traced hair and skin. The new Linear-Swept Sphere (LSS) data structure drastically improves performance and lowers memory usage for animated hair, delivering up to 2x faster rendering at just 20% of the VRAM cost compared to DOTS.

RTX Texture Filtering (RTXTF): A newly revealed feature that randomly samples textures post-shading to reduce artifacts and improve fidelity in tricky volume scenarios.

Streamline SDK Update

NVIDIA has also updated its Streamline SDK to v2.7.2, accessible via GitHub. This release adds support for:

• DLSS 4 Multi Frame Generation

• New models for Ray Reconstruction, Frame Generation, and Super Resolution

With the updated Streamline, developers and modders may soon implement DLSS 4 in their projects or community fixes, potentially expanding the reach of advanced NVIDIA AI features in both new and existing titles.

Do you think AI-accelerated features like RTX Neural Shaders and AI-based texture compression will significantly reshape how games are designed, or are they mostly incremental improvements? Share your perspective!