TurboQuant can reduce vector index size by 10x at 100M Row Scale
A new pull request for pgvector introduces TurboQuant, a quantization technique that can reduce PostgreSQL vector index size by up to 10x at the 100 million row scale, aiming to improve storage efficiency for large-scale vector search.
Background
pgvector is a popular open-source extension that adds vector similarity search to PostgreSQL, widely used for AI/ML applications like semantic search and RAG (retrieval-augmented generation). This GitHub PR introduces "TurboQuant," a compression technique that shrinks vector index storage by ~10x at 100M-row scale. The practical significance: vector indexes are memory-hungry and expensive to maintain; a 10x reduction makes large-scale vector search far more feasible on existing hardware, lowering cost and complexity for production AI systems that rely on PostgreSQL.