1. PROJECT TITLE:
CUDA Based GPU Acceleration of Scalar Transport Equation
and Navier Stokes Solver for Incompressible Flows
Submitted by:
Murali Mohana Krishna Dandu
Department of Mechanical Engineering
Indian Institute of Technology Guwahati, INDIA.
Guiding Professor:
Prof. Chao-An Lin
Chairman, Power Mechanical Engineering
National Tsing Hua University, TAIWAN.
2. Abstract
The present work has been carried out at the Computational Fluid Dynamics Lab at the
Department of Power Mechanical Engineering, NTHU, Taiwan. Now a day, parallel
computing using GPGPU has become an efficient way to maximize the performance
throughput of CFD simulations. The paper first gives an introduction to GPU computing
using CUDA architecture. The Scalar Transport Equation has been solved using various
schemes like Upwind, CDS, QUICK and these are implemented on Cuda. The
Incompressible Navier Stokes solver has been implemented on gpu choosing the problem
of lid driven cavity flow. SIMPLE algorithm has been employed using a finite volume
staggered grid approach and various high Re flows have been compared with the
benchmark data. Codes for NS solver using Projection method are also been developed.
Finally, the performance of CPU and GPU has been compared. The speedup ratios of
NVIDIA Tesla C2050 and GeForce GTX Titan versus Intel 8 Core i7 processor have
been discussed by changing various parameters. The simulation results using Cuda based
parallel algorithms showed an acceleration of about 8-10 times.
Problem solving skills, patience and debugging are the essential things that I have
developed during the internship for arriving at optimized solutions. I have learnt various
CFD techniques like FD, FV and several discretization schemes. I got good hands on
experience in coding using Matlab, C and CUDA and developed a good understanding of
parallel computing.