NVIDIA/cuvs

What it solves

cuVS provides high-performance, GPU-accelerated implementations of algorithms for approximate nearest neighbors (ANN) and clustering. It simplifies the process of using NVIDIA GPUs to perform vector similarity search, which is essential for handling large-scale unstructured data and semantic search.

How it works

Built on the RAPIDS RAFT library, cuVS implements state-of-the-art algorithms (such as CAGRA) to build indices and perform searches across dense vectors. It provides APIs for Python, C++, C, and Rust, allowing developers to integrate these GPU-accelerated routines into databases, machine learning pipelines, or standalone applications.

Who it’s for

Developers and data scientists working with large-scale vector search, Generative AI (RAG), recommender systems, computer vision, and data mining tasks like k-NN graph construction or clustering (e.g., K-means, HDBSCAN).

Highlights

• Multi-language support: Native APIs for Python, C++, C, and Rust.
• High Performance: Optimized for fast index building, low-latency search, and high throughput.
• Broad Application: Supports semantic search for text, audio, image, and molecular data.
• Interoperability: Ability to build indices on GPU and deploy them on CPU.
• Automatic Optimization: Maintains updated code for new NVIDIA architectures and CUDA versions.