Parallel vertex clustering (CUDA)
Date
2024
Authors
Brolo, Enrique
Journal Title
Journal ISSN
Volume Title
Publisher
University of Victoria
Abstract
The computational demands of high-quality graphics are rising, and the need to manage these large datasets effectively is apparent. This project aims to rewrite and optimize a parallel vertex clustering algorithm by leveraging CUDA to accelerate the processing of large-scale 3D meshes. Aside from increasing processing speed, the project also looks at scalability and handling of massive datasets which can be constrained by single CPU-based methods. Though attempts at parallelism with OpenMP and multi-threaded CPU approaches have proven effective to a degree, they are still limited by the CPU’s lower capacity for handling numerous operations at once. CUDA instead utilises many GPU cores at once, enabling far greater parallelization for processing complex geometric data. Although the algorithm is still in development, early results suggest performance gains, however methods can still be further optimized. Future work will aim to refine the algorithm and proceed with more tests to ensure it can scale effectively as well as maintain accuracy.
Description
Keywords
CUDA, parallel computing, vertex clustering, mesh reduction