Presentation

P25 - Implementation of the Performance-Portable ICON Model
DescriptionThe ICON modeling framework is a unified numerical weather and climate model used for operational numerical weather prediction as well as low- and high-resolution climate projection. It utilizes the Message-Passing Interface (MPI) for domain decomposition and has been extensively optimized for OpenMP multi-threading. The model was recently adapted for GPU systems in order to run high-resolution multi-year climate simulations at 2.8 km on the CSCS Piz Daint heterogeneous supercomputer. Because of the low compute intensity of atmospheric model and the cost of data transfer between CPU and GPU, it is necessary that all components within the time integration be ported to the accelerator. For the dynamics, most physical parameterizations and infrastructure code OpenACC compiler directives are used. For the soil parameterization, a Fortran-based domain-specific language "CLAW-DSL" is utilized. We discuss the challenges associated with porting this large community code, as well as the run-time and I/O challenges of simulations on large-scale systems at 2.8 km horizontal resolution. We also present many of the optimizations implemented for GPUs and high-resolution simulations, such as asynchronous I/O, GPU-to-GPU communication, asynchronous execution of kernels, data compression on GPUs, as well as performance comparisons of the full model on CPU and GPU.
TimeTuesday, 6 July 202117:30 - 19:00 CEST
Location
Event Type
Poster