Tiger Data Says PRISM Builds 100-Million-Vector Index in 1 Hour 40 Minutes
Summary
Tiger Data releases pg_vectorsearch, an open-source PostgreSQL extension in alpha, and shifts most vector-search development to its PRISM index. The company says PRISM builds a 100-million-vector index in 1 hour 40 minutes on a 128 GB machine, while pgvector's HNSW took 42 hours on 512 GB. It requires PostgreSQL 18 or later.
Key Points
- Tiger Data says PRISM answers 99% recall queries in 21 ms on a 128 GB machine where pgvector's HNSW fails to build.
- Tiger Data reports PRISM builds a 1-million-vector index in about 3 minutes on a 4 GiB machine; HNSW takes about 21.7 hours.
- Tiger Data plans automatic incremental rebalancing so PRISM partitions stay optimal under heavy data churn, replacing manual steps.