Development of an Efficient Method for Combustion Simulations Using a Hybrid CPU/GPU Computer System

Lund Combustion Engineering

Lund

IT Teknik Natur

Introduction

LOGE AB develops software aimed at solving complex chemical systems. In general terms, the
problem involves solving systems of multi-dimensional non-linear differential equations. A par-
ticular problem, is that the differential equations are stiff since different chemical reactions occur
at different time-scales. Therefore, efficient methods for solving the system of non-linear differen-
tial equations are of importance. With today’s multi-core CPUs and GPUs it is also of importance
to use algorithms and methods which use the potential of the hardware to its maximum. The
question to be addressed in this diploma work pertains to how to use the CPUs and GPUs as
efficiently as possible in order to achieve the shortest possible simulation wall-clock time.

Proposed Work
This master’s thesis will consist of the following parts:
- Literature survey of hybrid CPU/GPU techniques used in computational physics/computational
chemistry
- Implementation of a suitable hybrid CPU/GPU strategy
- Evaluation and benchmarking of different applications and different reactive systems
- Written report


Desired Profile
- Master of science in computer science, engineering physics or engineering mathematics
- Specialisation towards embedded systems and/or software development
- Knowledge of MPI, OpenMP and general parallelisation strategies
- Numerical analysis, ideally including numerical linear algebra and solution methods for
differential equations
- Further mathematics and a genuine interest in problem solving

 

Applications must include CV, personal letter and academic record.

 


Sista ansökningsdag
2014-12-12
Genomförs
VT15
Kontakt
...
Telefon
...
Notera att sista ansökningsdag passerats