Advances in Web-Age Information Management

Volume 4016 of the series Lecture Notes in Computer Science pp 372-384

Scalable Clustering Using Graphics Processors

  • Feng CaoAffiliated withDept. of Computer Science and Engineering, Fudan University
  • , Anthony K. H. TungAffiliated withSchool of Computing, National University of Singapore
  • , Aoying ZhouAffiliated withDept. of Computer Science and Engineering, Fudan University

* Final gross prices may vary according to local VAT.

Get Access


We present new algorithms for scalable clustering using graphics processors. Our basic approach is based on k-means. By changing the order of determining object labels, and exploiting the high computational power and pipeline of graphics processing units (GPUs) for distance computing and comparison, we speed up the k-means algorithm substantially. We introduce two strategies for retrieving data from the GPU, taking into account the low bandwidth from the GPU back to the main memory. We also extend our GPU-based approach to data stream clustering. We implement our algorithms in a PC with a Pentium IV 3.4G CPU and a NVIDIA GeForce 6800 GT graphics card. Our comprehensive performance study shows that the common GPU in desktop computers could be an efficient co-processor of CPU in traditional and data stream clustering.