J59.1 Accelerating the Cloud Scheme Within the Unified Model for CPU-GPU Based High-Performance Computing Systems

Thursday, 16 January 2020: 8:30 AM
257AB (Boston Convention and Exhibition Center)
Wei Zhang, ORNL, Oak Ridge, TN; and M. Xu, M. Morales Hernandez, M. R. Norman, S. Mahajan, K. J. Evans, A. Hill, and B. Shipway

The newly developed Cloud-AeroSol Interacting Microphysics (CASIM) module in the Unified Model (UM) is a multi-moment, five hydrometeor classes bulk scheme. With more detailed representations of cloud microphysical processes and their interaction with the aerosol environments, CASIM increases the accuracy of cloud-aerosol simulations. Meanwhile, the improved functionality results in more intensive computational calculations, leading to a potential improvement using modern computing architectures such as OpenACC for Graphical Processing Units (GPUs). This research aims to accelerate CASIM by offloading the most intensive computations from the host CPU to GPUs on Oak Ridge Leadership Computing Facility’s (OLCF) Summit machine. Both manual refactoring and automatic source-to-source compiler (CLAW) are used to implement OpenACC directives. The performance of the original and accelerated CASIM are primarily compared based on warm-cloud simulations in the Kinematic Driver Model (KiD). The automatic GPU porting by CLAW is found to be useful as the guidance for manual refactoring. Future work will further improve the performance, and couple the accelerated CASIM to UM to benefit the full simulation of earth system.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner