KBest: Efficient Vector Search on Kunpeng CPU

Ch. 1. P. 2347–2356.

Ma K., Meiling W., Senkevich O., Zijian L., Daihao X., Малышев Д. С., Yangming L., Shihai X., Xiao Y., Radionov A., Weidi Z., Yuanzhan G., Zhiyu Z., Xin Y., Lin L., Junhao W., Yiding L., Yaoyao F., Gongyi W., Gong Z., Fei Y., Yingfan L.

Vector search, which returns the vectors most similar to a given query vector from a large vector dataset, underlies many important applications such as search, recommendation, and LLMs. To be economic, vector search needs to be efficient to reduce the resources required by a given query workload. However, existing vector search libraries (e.g., Faiss and DiskANN) are optimized for x86 CPU architectures (i.e., Intel and AMD CPUs) while Huawei
Kunpeng CPUs are based on the ARM architecture and competitive in compute power. In this paper, we present KBest as a vector search library tailored for the latest Kunpeng 920 CPUs. To be efficient, KBest incorporates extensive hardware-aware and algorithmic optimizations, which include single-instruction-multiple-data (SIMD)
accelerated distance computation, data prefetch, index refinement, early termination, and vector quantization. Experiment results show that KBest outperforms SOTA vector search libraries running on x86 CPUs, and our optimizations can improve the query throughput by over 2x. Currently, KBest serves applications from both our in-
ternal business and external enterprise clients with tens of millions of queries on a daily basis.

Ключевые слова: Vector Search, Kunpeng CPU, векторный поиск, Кунпенг ЦПУ

ПУБЛИКАЦИЯ ПОДГОТОВЛЕНА ПО РЕЗУЛЬТАТАМ ПРОЕКТА: Сетевые и графовые модели, сложность алгоритмов и интеллектуальный анализ данных (2025)

В книге

KDD '26: Proceedings of the 32th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

Association for Computing Machinery (ACM), 2026.