CAGRA: Highly Parallel Graph Construction and Approximate Nearest Neighbor Search for GPUs

TL;DR

CAGRA introduces a GPU-optimized graph construction and search algorithm for approximate nearest neighbor search, achieving significant speedups over existing CPU and GPU methods while maintaining high accuracy.

Contribution

The paper presents a novel parallel graph construction and search algorithm tailored for GPUs, significantly improving efficiency and throughput in ANNS tasks.

Findings

CAGRA is 2.2~27x faster in graph construction than HNSW.

In large-batch searches at 90-95% recall, CAGRA is 33~77x faster than HNSW.

For single queries at 95% recall, CAGRA is 3.4~53x faster than HNSW.

Abstract

Approximate Nearest Neighbor Search (ANNS) plays a critical role in various disciplines spanning data mining and artificial intelligence, from information retrieval and computer vision to natural language processing and recommender systems. Data volumes have soared in recent years and the computational cost of an exhaustive exact nearest neighbor search is often prohibitive, necessitating the adoption of approximate techniques. The balanced performance and recall of graph-based approaches have more recently garnered significant attention in ANNS algorithms, however, only a few studies have explored harnessing the power of GPUs and multi-core processors despite the widespread use of massively parallel and general-purpose computing. To bridge this gap, we introduce a novel parallel computing hardware-based proximity graph and search algorithm. By leveraging the high-performance capabilities of modern hardware, our approach achieves remarkable efficiency gains. In particular, our method surpasses existing CPU and GPU-based methods in constructing the proximity graph, demonstrating higher throughput in both large- and small-batch searches while maintaining compatible accuracy. In graph construction time, our method, CAGRA, is 2.2~27x faster than HNSW, which is one of the CPU SOTA implementations. In large-batch query throughput in the 90% to 95% recall range, our method is 33~77x faster than HNSW, and is 3.8~8.8x faster than the SOTA implementations for GPU. For a single query, our method is 3.4~53x faster than HNSW at 95% recall.