Supercomputers Cooling System Simulation Data for Deep Learning Surrogate Development: ExaDigiT's FMU-Generated Datasets for Summit, Marconi100, and Lassen Configurations

Overview This dataset provides comprehensive Functional Mock-up Unit (FMU)-generated simulation data for high-performance computing (HPC) datacenter cooling systems, supporting the development and benchmarking of deep learning surrogate models within the ExaDigiT digital twin framework. The dataset contains thermo-fluid simulation outputs from three supercomputer cooling configurations: Marconi100 (CINECA) Lassen (Lawrence Livermore National Laboratory) Summit (Oak Ridge National Laboratory) Purpose The data enables: Training and evaluation of deep learning surrogates (LSTM, DeepONet, DeepM&Mnet) Benchmarking alternative surrogate architectures Uncertainty quantification studies Physics-informed machine learning research Data Generation All data was generated using FMU-based cooling models derived from the Modelica TRANSFORM library, integrated within the ExaDigiT framework. The FMU simulator provides high-fidelity predictions across entire cooling infrastructures.