WRF (Weather Research and Forecasting) is an important benchmark for weather modeling, computational scientists, and procurements. WRF is a mesoscale numerical weather prediction system designed to serve both atmospheric research and operational forecasting needs. It allows researchers to generate atmospheric simulations based on real data (observations, analyses) or idealized conditions and has a large worldwide community of registered users (over 25,000 in over 130 countries). The June 24 presentation “Performance-related developments in WRF” by John Michalakes
(NOAA/EMC), Mike Iacono and David Berthiaume (AER) and Indraneil Gokhale (Intel Corp) notes “Neither accelerator (Phi or M2070) is living up to its name” and to “wait for Cori” the Intel Knights Landing Nersc Supercomputer.Unfortunately, this presentation is based on old versions of both the Intel and NVIDIA hardware (an M2070). In contrast, NVIDIA has been actively optimizing WRF and other atmospheric codes (see “Progress of GPU Parallel NWP and Climate Models” from GTC2014) with current hardware and using OpenACC along with CUDA.
Thanks to HPC_guru for bringing this performance comparison to our attention.
BTW: Love the concluding graphic!