P68 AceCAST GPU-enabled Weather Research and Forecasting model development and applications.
Elliott, Samuel, Paul Maravelias, Daniel Abdi, Christian Tenasescu, and Gene Pache, TempoQuest Inc.
The Weather
Research and Forecasting (WRF) model is an open source, mesoscale numerical
weather prediction system designed to serve both operational forecasting and
atmospheric research needs. TempoQuest (TQI) is translating WRF to run on
NVIDIA graphic processor units (GPUs.) The target acceleration goal is 10 times
speed up compared against the latest generation central processor units (CPUs)
on standard high performance computing clusters. TQI has achieved speedups of
up to 8 times.
The TQI GPU enabled WRF software is called AceCAST. AceCAST currently supports
a wide variety of physics configurations used by the WRF community. AceCAST is
currently built from WRF version 3.8.1 and TQI intends on supporting version
4.0 in the near future. TQI offers both a licensed version of AceCAST (via
container or binary executable) as well as an AWS-based scalable platform for
running AceCAST-WRF simulations on GPU cloud computing resources. This
presentation will inform WRF users about AceCAST applications, performance, support
and development plans as well as the AceCAST-WRF cloud-based modeling platform.