• Home
  • Categories
  • Tags
  • Pricing
  • Submit
  1. Home
  2. Sdks & Libraries
  3. RTNN

RTNN

RTNN is a research prototype system and codebase that accelerates high-dimensional nearest neighbor search using hardware ray tracing units on modern GPUs. It targets vector similarity search workloads common in AI applications, exploring ray-tracing hardware as an alternative acceleration path to traditional CPU- or CUDA-based ANN indexes.

🌐Visit Website

About this tool

RTNN

Overview

RTNN is a research prototype system and codebase that accelerates neighbor search by mapping it onto Nvidia GPU hardware ray tracing units (RT cores). It focuses on low-dimensional spaces (≤3D), as commonly found in engineering and scientific simulations (e.g., particles, surface samples in CFD, graphics, and vision). The implementation is primarily based on the OptiX ray tracing API with CUDA used for parallel helper routines.

  • Category: SDKs & Libraries
  • Use cases: Vector / spatial similarity search, neighbor search in 2D/3D simulations and analysis
  • Source: https://github.com/horizon-research/rtnn

Features

Neighbor Search Capabilities

  • Supports fixed-radius search (range search).
  • Supports K-nearest neighbor (KNN) search.
  • Interface assumes:
    • A search radius r for both range and KNN searches.
    • A maximum neighbor count K for both KNN and range searches.
  • Can emulate unbounded KNN by using a very large r.
  • Can emulate unbounded range search by using a very large K.
  • Designed for low-dimensional (2D/3D) spatial neighbor search, common in physical simulations and geometric data.

Performance & Acceleration

  • Uses Nvidia RT cores (hardware ray tracing units) to accelerate neighbor search.
  • Demonstrates order-of-magnitude speedups over traditional GPU (CUDA-only) neighbor search implementations, as reported in an associated PPoPP 2022 paper.
  • Explores ray tracing hardware as an alternative acceleration path to standard CPU/CUDA approximate nearest neighbor (ANN) indexes for similarity search workloads.

Implementation & Technology

  • Built primarily on Nvidia OptiX (ray tracing programming interface).
  • Uses CUDA to parallelize many non-ray-tracing helper functions.
  • Includes required OptiX SDK 7.1.0 headers directly in the repository (no separate SDK install required for those headers).
  • Repository structure:
    • include/ – OptiX headers (copied from OptiX SDK 7.1.0, unmodified).
    • src/ – Source code for RTNN implementation.

Design Considerations

  • Search radius r even in KNN:
    • Reflects practical need to discard far-away neighbors whose influence (e.g., physical forces) is negligible.
  • Bound on K even in range search:
    • Controls memory usage.
    • Conforms to downstream tasks that expect a fixed or bounded number of neighbors.

Research Context

  • Technical design and evaluation are discussed in a PPoPP 2022 paper: https://horizon-lab.org/pubs/ppopp22.pdf.
  • Performance has been significantly improved since the version described in the paper.

Requirements

  • Nvidia GPU with RT cores (supports hardware ray tracing).
  • OptiX 7.1.0 headers are included; external installation of the SDK is not strictly required for those headers.
  • CUDA environment for building and running helper kernels.

Pricing

  • Open-source library (license included in LICENSE file in the repository).
  • No pricing information is provided; usage is governed by the repository’s open-source license.
Surveys

Loading more......

Information

Websitegithub.com
PublishedDec 25, 2025

Categories

1 Item
Sdks & Libraries

Tags

3 Items
#GPU acceleration
#ANN
#similarity search

Similar Products

6 result(s)
GTS

GTS is a GPU-based tree index for fast similarity search over high-dimensional vector data, providing an efficient ANN index structure that can be integrated into or used to build high-performance vector database systems.

Cagra

Cagra provides highly parallel graph construction and approximate nearest neighbor search for GPUs, supporting large-scale vector database operations and efficient similarity search.

Neighbor

Ruby gem for approximate nearest neighbor search that can integrate with pgvector and other backends to power vector similarity search in Ruby applications.

iRangeGraph

iRangeGraph is an ANN indexing approach and accompanying implementation for range-filtering nearest neighbor search. It provides a specialized graph-based index that supports vector similarity search under range constraints, making it directly useful as a component or reference implementation for advanced vector database indexing and retrieval.

NSG

NSG is an approximate nearest neighbor search algorithm based on a sparse navigable graph structure designed for high-dimensional vector similarity search. The reference implementation provides a graph-based ANN index that can be integrated into custom vector retrieval systems.

PilotANN

PilotANN is a memory-bounded GPU-accelerated framework for large-scale vector search, designed to improve performance and efficiency of approximate nearest neighbor (ANN) search workloads, making it relevant as a high-performance engine/component in vector database and vector search systems.

Built with
Ever Works
Ever Works

Connect with us

Stay Updated

Get the latest updates and exclusive content delivered to your inbox.

Product

  • Categories
  • Tags
  • Pricing
  • Help

Clients

  • Sign In
  • Register
  • Forgot password?

Company

  • About Us
  • Admin
  • Sitemap

Resources

  • Blog
  • Submit
  • API Documentation
All product names, logos, and brands are the property of their respective owners. All company, product, and service names used in this repository, related repositories, and associated websites are for identification purposes only. The use of these names, logos, and brands does not imply endorsement, affiliation, or sponsorship. This directory may include content generated by artificial intelligence.
Copyright © 2025 Acme. All rights reserved.·Terms of Service·Privacy Policy·Cookies