Elastic’s DiskBBQ is redefining vector search by breaking memory barriers and making large-scale AI applications faster and more cost-efficient.
Elastic (NYSE: ESTC), the Search AI Company, has unveiled DiskBBQ, a breakthrough vector search algorithm in Elasticsearch that significantly reduces memory dependency while improving performance and cost efficiency for large-scale AI workloads.
Traditionally, vector databases have relied on Hierarchical Navigable Small Worlds (HNSW) — a graph-based technique known for accuracy and speed, but limited by its heavy reliance on in-memory data. This constraint often leads to high infrastructure costs and scalability issues. Elastic’s new DiskBBQ algorithm, now available in Elasticsearch 9.2, addresses this limitation by introducing a disk-based approach that minimizes RAM usage without compromising on query speed or accuracy.
Powered by Better Binary Quantization (BBQ), DiskBBQ compresses vectors efficiently and clusters them into compact partitions for selective disk reads. This innovation reduces spikes in retrieval time, optimizes system performance, and allows seamless data ingestion even at massive scales.
“As AI applications scale, traditional vector storage formats force them to choose between slow indexing or significant infrastructure costs required to overcome memory limitations,” said Ajay Nair, General Manager, Platform at Elastic. “DiskBBQ is a smarter, more scalable approach to high-performance vector search on very large datasets that accelerates both indexing and retrieval.”
In benchmark testing, DiskBBQ demonstrated exceptional efficiency — sustaining query latencies of approximately 15 milliseconds while operating in as little as 100 MB of memory. Unlike HNSW, which must hold entire graphs in RAM, DiskBBQ intelligently reads only relevant clusters from disk, eliminating the memory bottleneck and enabling Elastic to scale to datasets limited only by CPU and disk capacity.
With DiskBBQ, Elastic is positioning Elasticsearch as a preferred platform for next-generation AI workloads — offering enterprises a path toward cost-effective, high-performance vector search suited for applications in retrieval-augmented generation (RAG), recommendation systems, and semantic search.
DiskBBQ is currently available in technical preview in Elasticsearch Serverless.
