NUMERICAL SIMULATIONS FOR GPU AND CPU EFFICIENCY COMPARISON APPLIED TO AUTOGENOUS WELDING PROCESSES

Traditional CPU (Central Processing Unit) computing techniques
mostly follow serial methodologies. However, the approaches employed in computational
fluid mechanics and heat transfer have great potential for processing
parallelization. The present work exposes a brief efficiency comparison between
numerical solutions using chips with GPU (Graphics Processing Unit) and CPU
type architectures. The analysis performed was focused on the three-dimensional
simulation of an autogenous welding process with a moving heat source. The solution
was developed by the application of the Finite Volume Method (FVM).
The simulations performed suggested that GPU processing is computationally
more optimized and, consequently, less expensive when compared to the traditional
CPU processing techniques.