A Study of CUDA Implementation in the Finite Element Methods
碩士 === 國立中正大學 === 數學系應用數學研究所 === 105 === In this paper, we update the work in [2] in 2012 and discuss the progress of NVIDIA’s CUDA from 2012 to 2016, which includes the software CUDA-toolkit and the hardware of GPUs. The model problems considered for speedup performance are the Laplace equation and...
Main Author: | |
---|---|
Other Authors: | |
Format: | Others |
Language: | en_US |
Published: |
2017
|
Online Access: | http://ndltd.ncl.edu.tw/handle/22958960082572653260 |
Summary: | 碩士 === 國立中正大學 === 數學系應用數學研究所 === 105 === In this paper, we update the work in [2] in 2012 and discuss the progress of NVIDIA’s CUDA from 2012 to 2016, which includes the software CUDA-toolkit and the hardware of GPUs. The model problems considered for speedup performance are the Laplace equation and Stokes problem. One of the goal is to study the effect of CUDA programming of the conjugate gradient (CG) method used to solve the symmetric positive definite matrix in Laplace equation. For the Stokes problem, generalized minimal residual (GMRES) method is used to solve non-symmetric matrix. The performance of the CUDA programming will be presented. For more accuracy, double precision is also considered in our computations.
|
---|