Application background CUDA on the kMeans to achieve GPU, a cluster analysis.We often come into contact with the clustering analysis. It is a common practice to extract N features, and then put them together to form a N vector, so that a mapping from the original data set to a N dimensional vector space is needed.Can also be applied to image segmentation, etc.. Key Technology CUDA GPU implementation of kMeans, parallel computing, data division and other related technologies.The steps of the algorithm are as follows:1, from the D K random elements, as the K of the cluster of their respective centers.2, calculated the remaining elements to the centers K cluster dissimilarity, these elements will be classified to the lowest cluster dissimilarity.3. According to the results of clustering, the K cluster is calculated. The calculation method is the arithmetic mean of all the elements in the cluster.4, the D of all elements in accordance with the new ce