Efficient Multi-GPU K-means Clustering

Simple item page
dc.contributor.advisorBurtscher, Martin
dc.contributor.authorO'Ryan, Kian
dc.contributor.committeeMemberGao, Byron
dc.contributor.committeeMemberMetsis, Vangelis
dc.date.accessioned2023-04-21T18:26:44Z
dc.date.available2023-04-21T18:26:44Z
dc.date.issued2023-04
dc.description.abstractOne of the best ways of digesting ever-growing large troves of gathered data is by grouping, also known as clustering, similar data points based on their attributes. A successful clusterization is accomplished by minimizing intra-cluster and maximizing inter-cluster attribute variation. Although theoretically simple, clustering has many real-life uses in fields such as astronomy, data processing, medicine, digital marketing, biology, and computer vision. One of the standard algorithms for accomplishing clusterization is k-means, also known as Lloyd’s algorithm. Lloyd’s algorithm for computing clusters has a simple heuristic implementation, but due to the extreme size of the typical input target data, it can be a very time-consuming algorithm and is, therefore, a suitable parallelization candidate. I report my novel methods for significantly speeding up this algorithm for both GPU and CPU. This was accomplished by minimizing the amount of communication between the device and the host. My implementations utilized CUDA for the GPU code and relied on OpenMP for CPU parallelization. I achieved respectable speedup levels compared to the current state of art implementations by NVIDIA and Meta. My GPU code can execute k-means 3000 times faster than Meta’s parallel CPU and 55 times faster than NVIDIA’s GPU code.
dc.description.departmentComputer Science
dc.formatText
dc.format.extent45 pages
dc.format.medium1 file (.pdf)
dc.identifier.citationO'Ryan, K. (2023). Efficient multi-GPU K-means clustering (Unpublished thesis). Texas State University, San Marcos, Texas.
dc.identifier.urihttps://hdl.handle.net/10877/16628
dc.language.isoen
dc.subjectk-means
dc.subjectLoyd's algorithm
dc.subjectLloyd's clustering
dc.subjectGPU-optimization
dc.subjectGPU
dc.subjectCPU
dc.subjectOpenmp
dc.subjectparallel processing
dc.subjectparallel computing
dc.subjectefficient computing
dc.subjectECL
dc.subjectefficient computing lab
dc.subjectclustering
dc.subjectcluster analysis
dc.subjectunsupervised learning
dc.subjectheuristics
dc.titleEfficient Multi-GPU K-means Clustering
dc.typeThesis
thesis.degree.departmentComputer Science
thesis.degree.disciplineComputer Science
thesis.degree.grantorTexas State University
thesis.degree.levelMasters
thesis.degree.nameMaster of Science

Files

Original bundle

Now showing 1 - 1 of 1
Thumbnail Image
Name:
ORYAN-THESIS-2023.pdf
Size:
6.3 MB
Format:
Adobe Portable Document Format
Download

Collections

Graduate Theses and Dissertations