Dissertation > Excellent graduate degree dissertation topics show
Research of Finite Element Method on GPU
Author: HuYaoGuo
Tutor: YangWenBing
School: Huazhong University of Science and Technology
Course: Engineering Mechanics
Keywords: GPU CUDA Parallel computing High-performance computing Finite Element
CLC: O241.82
Type: Master's thesis
Year: 2011
Downloads: 76
Quote: 0
Read: Download Dissertation
Abstract
|
In recent years, with the continuous development of the graphics processor (GPU) hardware architecture of the GPU 's programmable performance continue to strengthen , at the same time a substantial increase in computing power , making the GPU began to appear in the field of numerical computation . With the constant improvement of the NVIDIA CUDA parallel computing platform GPU parallel computing continue to penetrate the various disciplines . Because of the particularity of the finite element method , the development is very slow GPU architecture mature before the transplant , so far only in the initial state . This article by analyzing the NVIDIA GPU architecture and programming model on CUDA platform characteristics , calculated in the conventional finite element analysis program than the larger assembly and sparse linear equations solving two parts transplanted to the GPU . Give full consideration to the CUDA platform for GPU hardware characteristics , selected conjugate gradient method (CG) iterative method of solving sparse linear equations , the method of calculation of the amount of the largest part of the sparse matrix vector multiplication ( SpMV ) . Taking into account the characteristics of the CUDA platform data read and write , first determine the total stiffness matrix of compressed storage format for the CSR , then reorder the nodes of the model , equation solver SpMV operation so valuable cache can better take advantage of the GPU . Order to ensure highly parallel computing in the assembly process , to prevent conflict in the data read and write , the model units are grouped such that each group of the unit adjacent to each other , so that a single each thread assembled calculation just in the total just in a position corresponding to not have any overlap , to prevent the potential of the data conflicts . CG method in vector operations using CUDA BLAS library instead of own SpMV operation , so that the equation solving also take full advantage of GPU resources . By calculating the 4 different types of unit model , compared with the calculation result of the CPU side , the GPU in the assembled part can be up to seven times acceleration , in the equation solving section highest reached 6.4 times the acceleration . The results show that the use of GPU resources finite element method can get better acceleration
|
Related Dissertations
- Research on the Mainframe System of Powder Mixed EDM and the Craft Experiment,TG661
- EDM Electrode Wear Mechanism and Control Research,TG661
- Research on the Structure and Characteristics of Permanent Magnetic Coupling,TH139
- Finite Element Analysis and Structure Optimal Design for Combined Seal at Normal Temperature and Low Temperature,TH136
- Simulation Analysis on Temperature Stress of RCC Arch Dam and Its Construction Joints Design Research,TV642.2
- Simplified Analysis of Semi-Rigid Steel Frames under Vertical Loading,TU391
- Based on Nonliner Finite Element Analysis of the Infilled Wall Frame Structure,TU323.5
- Stability Analysis of Roller Compacted Concrete Gravity Dam Based on Time-history Method,TV642.2
- Study on the Design and Biomechanical Behaviors of Individualized Artificial Knee Joint,R318.1
- Design of Testing Equipment for Electronic Products,TN06
- S-lay Installation Analysis for Deepwater Pipeline,TE973
- Research on Intelligent Aerial Bomb for Forest Fire Fighting,S762
- Multi-pin Internal Fixation of Calcaneal Fractures in Finite Element Analysis,R687.3
- The Analysis and Comparison of Adjacent Segment Degeneration of the Finite Element Caused by Three Posterior Lumbar Fusion Style,R687.3
- The Establishment and Preliminary Application of the Three-dimensional Finite Element Model of Ankle Joint,R687.3
- Visual Feedback and Memory Behavior Based GPU Parallel Ant Colony Algorithm,TP301.6
- Different Bottom Structure of Dental Brackets Affect the Adhesiveness-Three Dimensional Finite-element Analysis,R783.5
- Airborne radar antenna pedestal rapid design systems research and development,TN959.73
- Sham hose drum drive means and hydraulic system of the structure,TE973
- The magnetorheological damper mechanical properties and Gun Recoil,TB535.1
- Nonlinear Structural Dynamics Analysis and Optimization Design on Naval Gun Driver,TJ391
CLC: > Mathematical sciences and chemical > Mathematics > Computational Mathematics > Numerical Analysis > The numerical solution of differential equations, integral equations > Numerical Solution of Partial Differential Equations
© 2012 www.DissertationTopic.Net Mobile
|