AWS released a guide to choosing a vector store for Amazon Bedrock Knowledge Bases
AWS published a guide comparing three vector stores for Amazon Bedrock Knowledge Bases – OpenSearch, Aurora PostgreSQL with pgvector, and S3 Vectors – in terms of performance, costs, and suitability for specific RAG tasks.
AWS published a guide on its blog comparing three vector stores that can be used in the customer-managed configuration of Amazon Bedrock Knowledge Bases for RAG (Retrieval Augmented Generation) architectures: Amazon OpenSearch Service, Amazon Aurora PostgreSQL with the pgvector extension, and Amazon S3 Vectors. According to the article, RAG combines large language models with retrieval systems that convert a query into a vector and search a database for the semantically closest text passages to provide additional context for the model response.
According to the article, the stores differ in their capabilities: Amazon OpenSearch Service provides results from data held in memory, supports k-NN search and hybrid search combining keywords with vector search, and is available both as managed clusters and as OpenSearch Serverless. Amazon Aurora PostgreSQL with pgvector supports the IVFFlat and HNSW indexing methods, the L2, cosine, and inner product distance metrics, and vectors with up to 2000 dimensions. Amazon S3 Vectors is described as a cost-effective solution with search times under one second; according to AWS, it reduces vector storage costs by up to 90 percent compared with traditional vector databases.
Using product catalog search on an e-commerce platform as an example, the article recommends Amazon OpenSearch Serverless because, according to AWS, it handles vector search with single-digit millisecond latency and natively supports the filtering and aggregations needed for faceted navigation (e.g. by price, brand, or color). The article also mentions that the newer OpenSearch Serverless NextGen collections are not yet compatible with the Retrieve API of Amazon Bedrock Knowledge Bases, so the reported benchmarks were run on older Classic collections.
The available text of the source article is incomplete and does not cover the comparison for the other use-case scenarios described. Details can be found in the source article.
Why it matters
The guide gives developers concrete criteria for deciding which of the supported vector stores to use based on the type of RAG task, latency requirements, and budget, instead of choosing through trial and error. For companies running RAG applications on AWS, this offers an opportunity to reduce infrastructure costs (by up to 90 percent with S3 Vectors, according to AWS) or, alternatively, choose a higher-performance option where low latency is crucial, such as product catalog search.
Two audiences, two different impacts
What this means
For individuals
Developers building RAG applications on Amazon Bedrock Knowledge Bases get concrete guidance on which of the three supported vector stores to choose based on the type of task and latency and cost requirements.
For a business
The choice of vector store affects the cost and performance of RAG infrastructure built on AWS; according to AWS, Amazon S3 Vectors can reduce vector storage costs by up to 90 percent compared with traditional vector databases.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.