Concept
DiskANN is a graph-based approximate nearest neighbor search algorithm from Microsoft Research that keeps a compressed navigation layer in RAM and the full-precision graph on SSD. It is used in services such as Azure Cosmos DB and SQL Server vector search.
| Type | Concept |
|---|
2026-06-08 · Video on YouTube
The explosion of AI-powered applications has created a new challenge: how do you efficiently search through billions of vector embeddings without exploding infrastructure costs? In this episode, we explore the “billion-vector problem” and compare two leading…
HNSW works in memory with low latency and high recall but needs lots of RAM, while DiskANN uses SSDs to scale to massive datasets with less RAM. · From the episode: HNSW vs DiskANN for Vector Search in Azure AI…
DiskANN was originally developed by Microsoft Research. It is available in Azure AI Search as an alternative to HNSW for very large vector datasets. · From the episode: HNSW vs DiskANN for Vector Search in Azure AI…
DiskANN relies on SSDs rather than RAM, so memory costs fall for large datasets. This addresses the cost problem of searching through very large numbers of vector embeddings. · From the episode: HNSW vs DiskANN for Vector Search in Azure AI…
Short info above, community knowledge below: add practical tips, examples or corrections. New contributions are reviewed first; trusted members publish directly.
Nobody has listed this topic yet. Add it to your profile.
Maintained by Mirko Peters (Microsoft MVP) · Last updated:
EN
Deutsch
Español
Français
Italiano
Português
Nederlands
Polski
Bahasa Indonesia
Русский
Українська
العربية
עברית
日本語
한국어