How OpenSearch Service Boosts Vector Database Build Speed by Up to 10× and Cuts Costs by 75%

GPU Acceleration for Vector Indexing

The new GPU‑acceleration feature automatically detects vector‑indexing workloads when enabled on an Amazon OpenSearch Service domain or serverless collection. Users do not need to provision or manage GPU instances; billing is based on actual OpenSearch Compute Units (OCU) consumed.

To enable GPU acceleration, select the "Enable GPU acceleration" option in the console or run the following AWS CLI command:

$ aws opensearch-update-domain-config \
    --domain-name my-domain \
    --aiml-options '{"ServerlessVectorAcceleration": {"Enabled": true}}'

After enabling, create a vector index that supports remote index building:

PUT my-vector-index
{
  "settings": {
    "index.knn": true,
    "index.knn.remote_index_build.enabled": true
  },
  "mappings": {
    "properties": {
      "vector_field": {
        "type": "knn_vector",
        "dimension": 768
      },
      "text": {
        "type": "text"
      }
    }
  }
}

Bulk‑load vector data using the standard OpenSearch bulk API; the GPU‑accelerated workload is applied automatically during index construction and force‑merge operations:

POST my-vector-index/_bulk
{ "index": { "_id": "1" } }
{ "vector_field": [0.1,0.2,0.3,...], "text": "Sample document 1" }
{ "index": { "_id": "2" } }
{ "vector_field": [0.4,0.5,0.6,...], "text": "Sample document 2" }

Automatic Optimization of Vector Indexes

The automatic‑optimization feature lets users achieve the best trade‑off among retrieval latency, quality, and memory without deep vector‑search expertise. Compared with the default index configuration, it can further lower cost and improve recall, eliminating weeks‑long manual tuning.

Typical workflow for large‑scale vector ingestion:

Prepare a Parquet dataset in Amazon S3 for the target domain or collection.

Configure the index and enable automatic optimization (or configure manually for a single vector field).

Import data via the Vector Ingestion (Ingestion) menu; GPU acceleration is applied automatically, delivering up to a ten‑fold speed increase while costing only 25% of the original price.

During optimization, users can set latency targets (e.g., p90 = "Modest") and required recall (e.g., ≥ 0.9) to let the system generate suitable index settings.

Performance Benchmarks

AWS benchmark tests show that GPU acceleration improves index‑build speed by 6.4 × to 13.8 × across different workloads. End‑to‑end vector‑database construction can achieve up to ten‑fold faster build times and a 75 % cost reduction (to one‑quarter of the original cost).

Availability and Pricing

GPU acceleration is available in US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney, Tokyo), and Europe (Ireland). Automatic optimization is available in additional regions, including US East (Ohio), US West (Oregon), Asia Pacific (Mumbai, Singapore, Sydney, Tokyo), and Europe (Frankfurt, Ireland).

Pricing is based solely on the OCU resources actually consumed during vector‑index construction; see the Amazon OpenSearch Service pricing page for details.