Evaluation of HPC architectures for BRAMS numerical weather model

This paper investigates the performance of a weather forecasting application (Brazilian Regional Atmospheric Modeling System BRAMS) on a number of selected HPC clusters in order to understand the impact of different architectural configurations on its performance and scalability. We simulated atmosphere conditions over South America for 24 hours ahead with BRAMS, using 100 cores as a starting point (100 cores step). An extra set of executions took place from 10 to 100 cores (10 cores step) to identify more details about BRAMS performance. Results reveal differences in BRAMS performance and its relationship with interconnection (technology and topology). In conclusion, interconnection can limit application performance even with code improvement.