An Implementation Of K-Means Algorithm On GPU

Database mining techniques have been used by more and more companies fromthe huge amounts of data to identify with the company’s future development whichis most closely related to knowledge to assist decision makers to make the mostideal decision. As the traditional data mining algorithms, K-Means algorithm hasmany inadequacies in the computational, such as code amount is huge, slowconvergence and computing time-consuming. In recent years, the GPU (GraphicsProcessing Units) has high-speed development, the GPU is very suitable forlow-cost high-efficiency and high-performance parallel floating-pointcalculation, the GPGPU (General Purpose computation on Graphics Processing Units)is a graphics task which is professional graphics processor in the originalgeneral purpose computing tasks handled by the central processor. Modern GPUsare powerful parallel processing capabilities and programmable pipeline, makesit possible to use the stream processor to handle non-graphics data. So, if theK-Means algorithm ported to the GPGPU on, we could reduce the computation time.OpenCL (Open Computing Language) has its trademark rights by the Khronos workinggroup management, Apple is a platform-independent and truly heterogeneouscomputing resources in the sense of the solution and also is an important GPGPUsolutions.Speedup with the K-means algorithm will be based on the GPU OpenCL programmingmore effectively together, the better the K-Means algorithm proposed in thispaper based on the OpenCL parallel K-Means algorithm (OpenCL-based ParallelK-means, referred to as the OP the K-Means).Finally, according to the actual test case, the application of this article thedesign of parallel K-Means algorithm to cluster analysis and processing, asopposed to expensive multi-core CPU system, GPGPU architecture can reduce thepressure on business costs, greatly improve the speed of data mining operations.