Back to search

National Technical University of Athens (NTUA)

Αριθμητική επίλυση προβλημάτων αεροδυναμικής-αεροελαστικότητας σε επεξεργαστές καρτών γραφικών

Abstract

dc:description

The main objective of this PhD thesis was the development of software for the solution of large scale aerodynamic and aeroelastic problems, running on modern Graphics Processing Units (GPUs). The developed GPU-enabled software solves the 2D/3D steady/unsteady Navier-Stokes/Euler equations for compressible fluids on dynamic unstructured/hybrid grids. Different programming approaches were programmed and assessed in terms of parallel speed-up. These approaches can easily be implemented in any other GPU-enabled software solving general purpose PDEs on unstructured grids. A mixed precision arithmetic (MPA) solver has been proposed. The use of MPA (instead of double precision arithmetic, DPA) reduces the total number of global memory accesses and, thus, increases the parallel efficiency of the GPU software without harming the accuracy of the solution. Though, the vertex-centered approach of the finite volume technique on unstructured grids is used, the use of the developed GPU-solver running on a single GPU instead of a CPU-solver running on a single core of a modern CPU may decrease the computational cost up to 110 times depending on the grid size and the compute capabilities of the GPU used. This substantial decrease of the required computational cost affects positively both the aerodynamic/aeroelastic analysis of aerodynamic bodies and the optimization-design via evolutionary algorithms (EAs), where a large number of evaluations must be performed. Thus the use of GPUs extends the range of industrial applications that can be carried out using CFD analysis tools. The fast and less accurate SPA and the relatively slower and accurate MPA variants of the GPU-solver are used in an EA-based hierarchical optimization method which is suitable for GPUs. Due to memory limitations, a single GPU cannot be used for solving large scale aerodynamic/aeroelastic problems. For large scale problems, GPU clusters are employed. In the present thesis, a single CPU-thread was used to control the available GPUs per computational node. The communication between the GPUs of the interconnected nodes is based on the MPI parallel protocol.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2012

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Trompoukis, Xenofon
  • Τρομπούκης, Ξενοφών

Subjects

dc:subject × 15

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/28180

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Trompoukis, Xenofon; Τρομπούκης, Ξενοφών. Αριθμητική επίλυση προβλημάτων αεροδυναμικής-αεροελαστικότητας σε επεξεργαστές καρτών γραφικών. National Technical University of Athens (NTUA), 2012. http://hdl.handle.net/10442/hedi/28180