# cuVS

> NVIDIA RAPIDS cuVS is a GPU-accelerated library for vector search and clustering with CUDA-optimized HNSW, IVF, CAGRA, and PQ implementations. Supports batch queries for high QPS, suited for large-scale similarity search in GenAI apps. Achieves up to 12x faster indexing and lower latency vs CPU-only alternatives like FAISS CPU.

- **URL:** https://github.com/rapidsai/cuvs
- **Category:** GPU Accelerated Vector DBs
- **Tags:** Nvidia, Rapids, Cuda, Gpu Acceleration, Cagra
- **Updated:** 2026-04-23 19:19
- **Canonical page:** https://vectordb.ever.works/nl/items/cuvs

## Details

# cuVS

**Source:** [GitHub - rapidsai/cuvs](https://github.com/rapidsai/cuvs)

---

## Description

cuVS is an open-source library from RAPIDS for fast, GPU-accelerated vector search and clustering, designed to simplify and accelerate vector similarity search and clustering tasks using state-of-the-art algorithms on NVIDIA GPUs. It is derived from the RAPIDS RAFT library and will become its primary home for approximate nearest neighbors (ANN) and clustering algorithms in late 2024.

---

## Features

- **GPU-accelerated vector search:** High-performance implementations of vector similarity search algorithms on NVIDIA GPUs.
- **Approximate Nearest Neighbor (ANN) algorithms:** State-of-the-art ANN algorithms for scalable and efficient similarity search in high-dimensional spaces.
- **Clustering algorithms:** Algorithms for clustering, including support for methods like K-means and integration with algorithms used in t-SNE, UMAP, and HDBSCAN.
- **Integration-ready:** Can be used directly or integrated into databases and other libraries for fast vector search and clustering.
- **Modern CUDA support:** Built to leverage the latest NVIDIA architectures and CUDA versions, ensuring optimal performance and scalability.
- **Derived from RAPIDS RAFT:** Contains the most up-to-date ANN and clustering algorithms migrated from the RAFT library.
- **Enables graph construction:** Facilitates fast construction of nearest neighbor graphs, which can be used for further graph algorithms and analyses.
- **Open-source and actively maintained:** The library is open-source and will continue to receive updates and performance improvements as new hardware and CUDA versions are released.
- **Multi-language support:** Codebase includes support for Python, C++, Java, Go, and Rust bindings or interfaces.
- **Documentation and examples:** Includes documentation, examples, and notebooks to help users get started.

---

## Pricing

cuVS is open-source and free to use under its license.

---

## License

cuVS is released under an open-source license (see the repository for details).

---

_This Markdown mirror is generated by Ever Works for AI agents. The canonical HTML page is at https://vectordb.ever.works/nl/items/cuvs._