Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.

  

Experimental study of high speed particle ltering using a GPU

Author(s):

A. Botros
Conference/Journal:

Semester Thesis, HS12 (10209)
Abstract:

For real time computations, fast programs are crucial. One way to speed up the computations is to parallelise the code and let it run on parallel architectures. FPGAs, multicore CPUs and GPUs are some of the most often used structures. But these structures often require to learn new speci c and complex programming constructs such as CUDA or OpenMP. One way to speed up the computations with Matlab is to use Jacket, a tool for GPU computing with Matlab. Jacket was published in 2008 by accelereyes. In this semester thesis Jacket is evaluated in terms of speed up in comparison with normal CPU-computation. As a testing environment, particle lters were implemented which have a distinct parallel structure. Well-known bottlenecks of particle lters were analysed as well as runtime improvements over CPUs. Evaluation showed, that Jacket still has some teething troubles. Not all desirable combinations of Matlab functions are supported, yet. On top of that, a bug was detected which turns out to be a major drawback for the intended use of particle lters. Regardless of these aws, Jacket was still able to speed up the computations signi cantly, is easy to handle and intuitive to learn. This lead to the conclusion, that Jacket is indeed a suitable tool for simple, not too complicated applications but should not be used for applications where time is crucial or code complex.

Year:

2013
Type of Publication:

(13)Semester/Bachelor Thesis
Supervisor:

N. Kariotoglou

No Files for download available.
% Autogenerated BibTeX entry
@PhdThesis { Xxx:2013:IFA_4427
}
Permanent link