Author

Thomas Meyer

Date of Award

2013

First Advisor

Eric Kramer

Second Advisor

Clark Musselman

Abstract

The N-DOP (Nitrogen and Dissolved Organic Phosphorus) model is a biogeochemical model used to determine the flux of carbon through the various layers of the ocean. More broadly, the model is used to study a phenomenon known as the ocean’s biological pump. This phenomenon helps to regulate the amount of atmospheric CO2 and therefore plays an important role in climate change. This thesis documents the process of simulating the N-DOP model on a parallel computer architecture known as a graphics processing unit (GPU). Originally, the N-DOP model was simulated on a cluster of desktop computers. However, due the to complexity of the simulation, it took days, even weeks, to run on this cluster. GPUs, then, were explored as a means to increase the performance of the simulation. To run the simulation on the GPU, certain aspects of the code base written for the desktop cluster had to be reworked. The steps involved in porting the code are described in detail in this thesis. All programming was done in mix of CUDA C and CUDA Fortran, dialects of C and Fortran used with GPUs. At the conclusion of this thesis, the performance of the ported code is analyzed. Some ideas about increasing the ported code’s performance are also presented.

Share

COinS