NVIDIA: Open Source Ecosystem

Chapter 2. cuVS — Search for the Era of Unstructured Data

A Note for Early Release Readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 9th chapter of the final book. Please note that the GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

Most of the data produced today is unstructured: text, images, audio, video, code. The way modern systems make this data searchable is to embed it, representing each item as a dense vector such that similar items land near each other. Once everything is a vector, the core operation becomes nearest-neighbour search: given a query vector, find the most similar vectors in a collection that may hold billions of them. Over the past decade this “vector search” has become a foundational primitive, underpinning search engines, recommender systems, retrieval-augmented generation, deduplication, anomaly detection, and more.

cuVS (pronounced “coo-V-S”) is NVIDIA’s library for doing this on GPUs. It provides GPU-accelerated implementations of vector search, both exact and approximate, along with the supporting operations those workloads need: clustering, pairwise distance computation, and quantization. Clustering earns its place in that list twice ...