<ul><li>Researchers have proposed a hybrid CPU-GPU system called PilotANN for graph-based Approximate Nearest Neighbor Search (ANNS).</li><li>PilotANN addresses the limitations of existing ANNS implementations by utilizing both CPU and GPU resources.</li><li>The system employs a three-stage graph traversal process, combining GPU-accelerated subgraph traversal, CPU refinement, and precise search with complete vectors.</li><li>Experimental results show that PilotANN achieves significant speedups and cost-effectiveness compared to CPU-only approaches, making high-performance ANNS more accessible on common hardware configurations.</li></ul>

PilotANN: A Hybrid CPU-GPU System For Graph-based ANNS

Discover more