# Neptune Database vs Neptune Analytics — Decision Matrix

Verified against AWS Neptune Analytics docs and pricing page as of **2026-07-19**.

| Question | Prefer Neptune Database | Prefer Neptune Analytics |
| -------- | ----------------------- | ------------------------ |
| Need transactional writes? | Yes | No (read-only analytics graph) |
| Freshness requirement | Seconds (live cluster) | Snapshot / load time acceptable |
| Workload | Point traversals, CRUD | Full-graph algorithms (WCC, PageRank, community detection) |
| Vector + graph in one query | Limited / not primary path | Native vector index (up to 65,000 dims) + algorithms |
| Cost control between runs | Always-on instance or Serverless NCUs | Pause graph (≈10% compute) or delete + reload from S3 |
| Query language | Gremlin, openCypher, SPARQL | openCypher only |
| GraphRAG with Bedrock | Pair with KB elsewhere | Supported GraphRAG path via Neptune Analytics |

## Starting capacity (m-NCU)

1 m-NCU ≈ 1 GB memory + matching compute. Smaller **32** and **64** m-NCU sizes exist for pilots. Account max commonly defaults near **1024** m-NCU (verify in console). Size from actual import footprint — do not assume a hard 64 GB ceiling.

## Cost sketch (illustrative)

AWS pricing example (US East): 256 m-NCU × 2 hours ≈ **$15.36** ⇒ roughly **$0.03 / m-NCU-hour**. Re-check [Amazon Neptune pricing](https://aws.amazon.com/neptune/pricing/) before budgeting.