Optimizing NVIDIA CAGRA in Milvus: A Hybrid GPU–CPU Approach to Faster Indexing and Cheaper Queries
Dec 10, 2025Marcelo Chen
Summary
Introduces and details the hybrid GPU-CPU approach for CAGRA indexing in Milvus 2.6.1. Explains the CAGRA algorithm (NN-Descent and 2-hop pruning) and its GPU acceleration. Details the `adapt_for_cpu` parameter and its four execution modes for balancing GPU construction with CPU querying. Presents experimental results demonstrating significant index build time improvements (12-15x) and discusses query performance and recall accuracy.
More from the AI Data Storage Engine trail
- Milvus Snapshots: Point-in-Time Collection Views Without Copying Data
- Announcing Milvus 3.0: Lake-Native Vector Search and a More Powerful Retrieval Engine
- How to Recover Milvus Data After etcd Corruption Without a Backup
- Open Tag: An Open-Source Claude Tag for Claude Code and Codex
- How Force Merge Compaction Nearly Doubled Milvus Search QPS