Date of Award
12-2019
Document Type
Dissertation - Campus Access Only
Degree Name
Doctor of Philosophy (PhD)
Degree Discipline
Electrical Engineering
Abstract
Transistors have been increasing the graphics processing unit’s (GPU) performance since GPUs were first used to conduct scientific research. Currently, Moore’s law is becoming less relevant, and transistors are no longer able to bring a notable performance boost to the GPU. This study outlines different ways that the GPU’s performance can be improved without depending on transistors. In particular, this study focuses on problems associated with the GPU and proposes solutions that help correct those problems. When the GPU is executing dynamic parallelism and fine grain synchronization applications, the GPU performance can decline. To help improve the GPU execute those applications, this study introduces a dynamic parallelism management design and a fine grain synchronization management design. The dynamic parallelism management design and the fine grain synchronization management design are able to bring a 39% and 26% speedup performance improvement for the GPU, respectively. This study also explores a non-uniform memory access (NUMA) based GPU, which is a post Moore’s law era GPU. However, the NUMA based GPU has a problem serving memory requests, which causes it not to reach its full performance potential. To help the NUMA based GPU better serve memory requests, this study introduces several units for the NUMA based GPU. By implementing the units to the NUMA based GPU, the execution time improved by 32%.
Committee Chair/Advisor
Pamela Obiomon
Committee Member
Matthew Sadiku
Committee Member
Charlie Tolliver
Committee Member
Xishuang Dong
Committee Member
Sherri Frizell
Publisher
Prairie View A&M University
Rights
© 2021 Prairie View A & M University
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
Date of Digitization
11/13/2024
Contributing Institution
John B Coleman Library
City of Publication
Prairie View
MIME Type
Application/PDF
Recommended Citation
Bourgeois, D. J. (2019). Graphics Processing Unit (GPU) Performance Optimization: Post Moore’s Law Era. Retrieved from https://digitalcommons.pvamu.edu/pvamu-dissertations/89