|
|
 |
|
Accuracy analysis of a lattice Boltzmann method with fixed-point arithmetic with focus on a FPGA implementation
Supervision:
Background:
FPGAs are of emerging interest for high-performance applications. FPGAs are freely programmable hardware devices that originally are
used for rapid prototyping in the design of microelectronic curcuits. Due to the arbitrary combinations of computing elements that
can be configured on FPGAs, they are also of interest for scientific problems with a need for massive computations. However, floating
point operations needed for most of scientific applications are very hard to implement on FPGAs due to timing and area reasons.
Mostly one reduces the accuracy of computations in order to make an implementation on FPGAs feasible.
Tasks:
The topic of this thesis is to examine the lattice Boltzmann method with regard to its need for computational accuracy. The goal
is to find a range of parameters within which this method for computational fluid dynamics remains valid, while the computations are
performed with reduced accuracy such as single-precision, different widths of operands or even fixed-point arithmetic.
If the student is interested, he has the possibility to implement and test the method using a SGI RASC RC100. These blades are
part of modern Altix supercomputers and feature two Xilinx Virtex FPGAs each. For such an implementaion excellent programming skills
as well as good command in at least one hardware description language are required.
Requirements:
- Skills in numerical mathematics
- Skills in hardware, esp. FPGAs, Signals, Gates, etc. (preferably study of microelectronics or similar)
- Programming skills
Type:
Master Thesis or Diploma Thesis
Status:
Free
|
|
|
|
