An efficient parallel collaborative filtering algorithm on multi-GPU platform

Collaborative filtering (CF) is one of the essential algorithms in recommendation system. As the size of the data in real applications is huge, usually at the magnitude of Petabytes, parallel computing technique is required to accelerate the computation. Due to GPU’s tremendous computing capability, it has emerged as the co-processor of the CPU to achieve a high overall throughput. In this paper, we identify the computation kernel, similarity matrix calculation. Then, we present a CUDA multithread model, where the data elements are processed in a data-parallel fashion. A workload partitioning scheme is proposed to balance the workload distributed to different GPUs. In the experiments, our CUDA-enabled CF algorithm significantly outperforms the serial CF workstation, achieving up to 3,691×\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times $$\end{document} speedup using four Tesla K10 graphics cards. It also shows good scalability when varying the number of users, the number of items and the number of GPUs.