Abstract
Abstract. The best hope for reducing long-standing global climate model biases is by increasing resolution to the kilometer scale. Here we present results from an ultrahigh-resolution non-hydrostatic climate model for a near-global setup running on the full Piz Daint supercomputer on 4888 GPUs (graphics processing units). The dynamical core of the model has been completely rewritten using a domain-specific language (DSL) for performance portability across different hardware architectures. Physical parameterizations and diagnostics have been ported using compiler directives. To our knowledge this represents the first complete atmospheric model being run entirely on accelerators on this scale. At a grid spacing of 930 m (1.9 km), we achieve a simulation throughput of 0.043 (0.23) simulated years per day and an energy consumption of 596 MWh per simulated year. Furthermore, we propose a new memory usage efficiency (MUE) metric that considers how efficiently the memory bandwidth – the dominant bottleneck of climate codes – is being used.
Highlights
Should global warming occur at the upper end of the range of current projections, the local impacts of unmitigated climate change would be dramatic
To enable the running of Consortium for Small-Scale Modeling (COSMO) on hybrid highperformance computing systems with graphics processing units (GPUs)-accelerated compute nodes, we rewrote the dynamical core of the model, which implements the solution to the non-hydrostatic Euler equations, from Fortran to C++ (Fuhrer et al, 2014)
Our implementation of the COSMO model that is used for production-level numerical weather predictions at MeteoSwiss has been scaled to the full system on 4888 nodes of Piz Daint, a GPU-accelerated Cray XC50 supercomputer at the Swiss National Supercomputing Centre (CSCS)
Summary
Should global warming occur at the upper end of the range of current projections, the local impacts of unmitigated climate change would be dramatic. For Prediction Across Scales (MPAS) and later, in 2015, participated in the Generation Global Prediction System (NGGPS) model intercomparison project (Michalakes et al, 2015) at the same resolution and achieved 0.16 SYPD on the full National Energy Research Scientific Computing Center (NERSC) Edison system. Yang et al (2016a) use an acoustic Courant number up to 177; i.e., their time step is 177 times larger than in a standard explicit integration (this estimate is based on the x = 488 m simulation with t = 240 s) In their case, such a large time step may be chosen, as the sound propagation is not relevant for weather phenomena. Since the IFS model is not a non-hydrostatic model, we conclude that even for fully implicit, global, convection-resolving climate simulations at ∼ 1–2 km grid spacing, a time step larger than 40–60 s cannot be considered a viable option. The main advantage of this approach is that it exhibits – at least in theory – perfect weak scaling. This applies to the communication load per sub-domain, when applying horizontal domain decomposition
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
