Browsing by Author "Ples, Vladimir"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Implementation of a model design framework in a GPU Server PlatformPublication . Ples, Vladimir; Ruano, AntónioThis thesis deals with the development and implementation of a model design framework in a Graphical Processing Unit (GPU) Server Platform powered by a multi-objective genetic algorithm with some novel components. With GPUs suffering rapid improvements in performance, functionality and energy efficiency they have become an attractive choice in contexts where it is necessary to process large amounts of data. The objective behind this work is to make use of the processing capabilities of the GPU server platform to accelerate the process of model design by parallelizing the workload in the GPU Server Platform and making use of its specific capabilities to provably outperform an existing legacy platform using a cluster of machines that rely mostly on Central Processing Units (CPUs). This thesis will first provide a background on the GPU and the Multi-Objective Genetic Algorithm that will be used, as well as the Legacy Platform that will be employed as a basis. The following section will provide an analysis of the state of the art regarding constructive algorithms and the general paradigm in which GPUs are used. The next sections details the initial contact with the legacy model design framework and the documentation efforts; the creation of a methodology to extract processing times and memory values; the prototyping of the GPU implementation along with the challenges presented, and the solutions found. The final parts of the thesis address the statistical validation of the newly produced models as well as a comparison of the performance between the legacy platform and the new one.