Data for TChem-atm (v2.0.0): Scalable Performance-Portable Multiphase Atmospheric Chemistry
Abstract
This dataset contains the following to replicate figures from "TChem-atm (v2.0.0): Scalable Performance-Portable Multiphase Atmospheric Chemistry" submitted to Geophysical Model Development (GMD). It contains (1) the simulation inputs, outputs and analysis notebook for recreating the PartMC-CAMP and PartMC-TChem-atm comparison and (2) scripts, timing results and analysis tools for recreating the performance evaluation. Users can either inspect the raw output to verify the results of the manuscript or rerun simulations using the provided inputs. Additionally, modifiying the inputs allows for for further exploration of both model simulation and performance characteristics.
Full text
TChem-atm (v2.0): Scalable Performance-Portable Multiphase Atmospheric Chemistry Oscar H. Díaz-Ibarra1, Samuel G. Frederick2, Jeffrey H. Curtis2, Zachary D’Aquino2, Peter A. Bosler1, Lekha Patel1, Cosmin Safta3, Matthew West4, and Nicole Riemer2 1Center for Computing Research, Sandia National Laboratories, Albuquerque, NM, USA 2Department of Climate, Meteorology, and Atmospheric Sciences University of Illinois Urbana-Champaign, Urbana, IL, USA 3Data Sciences and Computing, Sandia National Laboratories, Livermore, CA, USA 4Department of Mechanical Science and Engineering, University of Illinois Urbana-Champaign, Urbana, IL, USA This dataset contains all material required to produce the figures found within the manuscript submitted to Geoscientific Model Development entitled “TChem-atm (v2.0): Scalable Performance-Portable Multiphase Atmospheric Chemistry“. The archived dataset consists of two downloadable files: –partmc_simulations.tar.gz: PartMC-TChem-atm and PartMC-CAMP data from Section 3.1 –timings.tar.gz: TChem-atm CPU/GPU timing experiments from Section 3.25 Software requirements All figures in the paper were produced with Python 3.12.9. Required packages are as follows with the version used for this manuscript: –pandas (2.2.3) –json (2.0.9)10 –numpy (2.2.3) –matplotlib (3.10.0) –netCDF4 (1.7.2) - available at https://unidata.github.io/netcdf4-python/ Directory structure of archived simulation data for Section 3.1 Upon downloading and untarring partmc_simulations.tar.gz, it may be explored as follows:15 –Input data files for conducting the set of three simulations for the CB05+SIMPOL mechanism in scenarios/ –camp.spec and relevant input files specified within for running PartMC-CAMP 1
–tchem.spec and relevant input files specified within for running PartMC-TChem-atm on CPU –tchem_gpu.spec and relevant input files specified within for running PartMC-TChem-atm on a GPU –Output data in the out/ directory consists of netCDF files per output time:20 –CAMP output: camp_0001_* –PartMC-TChem-atm CPU output: tchem_cb05cl_ae5_* –PartMC-TChem-atm GPU output: tchem_gpu_cb05cl_ae5_* –simulation_notebook.ipynb: Python Jupyter notebook for producing Fig. 5 and analysis of error. Directory structure of archived numerical experiment data for Section 3.225 Upon downloading and untarring timings.tar.gz, it may be explored as follows: –data/: directory containing all the timing results. The descriptions of the individual directories are described in Table 1. –scripts/: directory containing the following scripts: –solver_plots.ipynb: Python Jupyter notebook for producing Fig. 6, 7. –rhs_plots.ipynb: Python Jupyter notebook for producing Fig. 8.30 Cluster Experiment Platform Path relative to data directory DeltaAI RHS NVIDIA H100 GPU deltaAI/CB05CL_AE5_w_simpolSOA/CUDA/rhs-no_sacado DeltaAI Jacobian NVIDIA H100 GPU deltaAI/CB05CL_AE5_w_simpolSOA/CUDA/rhss-no_sacado DeltaAI TrBDF2 NVIDIA H100 GPU deltaAI/CB05CL_AE5_w_simpolSOA/CUDA/trbdf-no_sacado DeltaAI Sundials CVODE NVIDIA H100 GPU deltaAI/CB05CL_AE5_w_simpolSOA/CUDA/sundials_dense-no_sacado DeltaAI Sundials CVODE-GMRES NVIDIA H100 GPU deltaAI/CB05CL_AE5_w_simpolSOA/CUDA/sundials_gmres-no_sacado Frontier RHS AMD MI250X GPU frontier/CB05CL_AE5_w_simpolSOA/HIP/rhs-no_sacado Frontier Jacobian AMD MI250X GPU frontier/CB05CL_AE5_w_simpolSOA/HIP/jac-no_sacado Frontier TrBDF2 AMD MI250X GPU frontier/CB05CL_AE5_w_simpolSOA/HIP/trbdf-no_sacado Frontier Sundials CVODE AMD MI250X GPU frontier/CB05CL_AE5_w_simpolSOA/HIP/sundials_dense-no_sacado Frontier Sundials CVODE-GMRES AMD MI250X GPU frontier/CB05CL_AE5_w_simpolSOA/HIP/sundials_gmres-no_sacado Perlmutter RHS AMD 7763 CPU perlmutter/CB05CL_AE5_w_simpolSOA/HOST/rhs-no_sacado Perlmutter Jacobian AMD 7763 CPU perlmutter/CB05CL_AE5_w_simpolSOA/HOST/jac-no_sacado Perlmutter TrBDF2 AMD 7763 CPU perlmutter/CB05CL_AE5_w_simpolSOA/HOST/trbdf-no_sacado Perlmutter Sundials CVODE AMD 7763 CPU perlmutter/CB05CL_AE5_w_simpolSOA/HOST/sundials_dense-no_sacado Perlmutter Sundials CVODE-GMRES AMD 7763 CPU perlmutter/CB05CL_AE5_w_simpolSOA/HOST/sundials_gmres-no_sacado Table 1. Experiment configurations across different clusters and platforms 2