NVIDIA-RTX/NRD

NVIDIA Real-time Denoising (NRD) library

What it solves

NRD provides a high-quality, real-time denoising library for noisy signals, specifically targeting 1 path/pixel path tracing. It resolves noise on opaque surfaces for both static and dynamic lighting, offering a non-AI alternative to technologies like DLSS-RR that maintains high visual fidelity in real-time rendering.

How it works

NRD is an API-agnostic, spatio-temporal library that uses per-pixel G-buffer guides (including normals, roughness, viewZ, and motion vectors) to clean up noisy radiance and occlusion signals. It includes three specialized denoisers:

  • REBLUR: A recurrent blur-based denoiser for diffuse and specular radiance, as well as occlusion.
  • RELAX: An A-trous based denoiser designed for RTX Direct Illumination (RTXDI), handling diffuse and specular radiance.
  • SIGMA: A per-light shadow-only denoiser for infinite or local light sources, including translucency.

Who it’s for

It is designed for developers of AAA game titles and professional visualization (ProVis) applications who need to integrate high-performance denoising into D3D12, Vulkan, or D3D11 engines.

Highlights

  • API Agnostic: Compatible with D3D12, Vulkan, and D3D11.
  • High Performance: Optimized for RTX hardware (e.g., achieving ~2.55ms for REBLUR_DIFFUSE_SPECULAR on an RTX 4080 at 1440p).
  • Flexible Integration: Can be integrated via a high-level NRDIntegration layer using NRI (NVIDIA Rendering Interface) or via a native API implementation.
  • Versatile Signal Support: Handles diffuse/specular radiance, ambient/directional occlusion, and shadows.

Related

  • Project
    RenderKit/oidn

    A high-performance library of deep learning-based denoising filters that reduces rendering times for ray tracing applications by removing Monte Carlo noise.

  • Project
    perseval-BLR/NeuralScreen

    NeuralScreen is a Windows‑only overlay that runs NVIDIA’s DLSS 5 neural renderer on the whole desktop (or a single window) in real time, optionally adding DLSS‑G frame generation. It works with RTX GPUs, offers hot‑key control, per‑window mode, recording, and a configurable UI.

  • Project
    maanHimself/OpenDLSS-NR

    OpenDLSS‑NR is a Windows‑only Vulkan re‑implementation of NVIDIA’s DLSS 5 Neural Rendering network. It reproduces the exact 71‑block Swin‑ViT model (FP8 on tensor cores) and provides both a high‑performance PTX kernel path and a reference GLSL path, plus a Filament demo and a WebGPU browser version. Users supply their own model weights; the tool can benchmark, verify bit‑exactness against NVIDIA fixtures, and run interactive demos.

  • Project
    georghess/neurad-studio

    neurad‑studio is a research‑grade library that brings neural‑rendering (NeRF and 3‑D Gaussian splatting) to autonomous‑driving data. It supplies ready‑to‑run implementations of NeuRAD and SplatAD, parsers for major AD datasets, a viewer that can display both camera images and lidar points, and standard training/visualisation tools built on Nerfstudio.

  • Project
    CVCUDA/CV-CUDA

    CV‑CUDA is NVIDIA’s open‑source GPU‑accelerated computer‑vision library. It provides fast, scalable image/video operators (resize, filter, color conversion, etc.) that run entirely on the GPU and integrate with C/C++ and Python AI frameworks. Install via pre‑built wheels (`cvcuda-cu12` or `cvcuda-cu13`) for Linux, decode images directly to GPU with `nvimgcodec`, and chain CV‑CUDA operators before feeding data to deep‑learning models. The library targets modern NVIDIA GPUs (Turing and newer) and is used in production systems such as Microsoft Bing visual search and Tencent Cloud video processing.