P1 GPU acceleration of the WRF model with OpenACC

Posey, Stan, and Carl Ponder, NVIDIA

During 2012 a software standard named OpenACC was introduced for development of GPU acceleration through a directives-based approach with automatic parallelization. Also in 2012 NVIDIA made available an extensive set of new HPC architectural features in the company’s fourth generation of GPUs codenamed Kepler. Together these developments provided new opportunities for application development and performance gains in earth system modeling. This talk will examine the GPU progress of the WRF model using OpenACC directives along a review of the necessary programming procedures. The talk will also present examples of physics schemes that demonstrate substantial performance gains from GPU acceleration, and discuss requirements of OpenACC applied for the full WRF model.