# Product Quantization Compression

> Lossy vector compression dividing vectors into subvectors for independent quantization. Achieves 8-64x storage reduction while enabling fast approximate distance computation via lookup tables.

- **URL:** https://cybergarden.au/blog/5-powerful-vector-database-tools-2025
- **Category:** Concepts & Definitions
- **Tags:** compression, Quantization, pq
- **Updated:** 2026-03-11 12:14
- **Canonical page:** https://vectordb.ever.works/items/product-quantization-compression

## Details

## Overview

Product Quantization (PQ) compresses vectors by splitting them into subvectors and quantizing each segment independently, dramatically reducing memory footprint.

## Process

1. Divide vector into m subvectors
2. Cluster each subvector space (k-means)
3. Replace subvectors with centroid IDs
4. Store compressed codes + codebooks
5. Approximate distances using precomputed tables

## Compression

- **8-64x reduction** typical
- Configurable via m and k parameters
- Tradeoff with accuracy

## Variants

- **OPQ**: Optimized with rotation
- **IVFPQ**: Combined with IVF
- **HNSW-PQ**: Graph index with compression

## Used In

- FAISS
- Milvus
- ScaNN
- LanceDB

## Pricing

Open technique.

---

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