Optimization and performances testing of CUDA-(multi)GPU-accelerated codes for the automatic parameterization of physical models based on SDE Monte Carlo algorithms
The SDEGnO project aims to develop high-performance general-purpose simulators for the simulation and automatic calibration of physical models based on stochastic differential equations (SDEs). The initiative seeks to implement an integrated framework that, through the use of Monte Carlo techniques and global optimization algorithms (including evolutionary strategies and swarm intelligence methodologies), enables a significant reduction in simulation time and energy consumption, while ensuring high accuracy in results.
In the first phase, the team focused on optimizing CUDA code, fully leveraging the capabilities of NVIDIA multi-GPU architectures. This includes refining memory management and adopting Single Instruction, Multiple Data (SIMD) techniques to maximize computational efficiency. The next phase involves integrating advanced algorithms for the automatic search and calibration of physical parameters, allowing dynamic adaptation to various application scenarios. We will also develop new algorithms for the assessment of uncertainty and sensitivity of parameters.
Prof. Marco S. Nobile; Leone Bacciu; Dr. Giovanni Cavallotto; Prof. Sabina Rossi; Prof. Michele Bugliesi (guest); Dr. Stefano Dalla Torre; Matteo Grazioso.
The resulting framework, replicable and scalable, will provide strategic support to future research groups in the HPC domain and contribute to strengthening the scientific competitiveness and economic impact of the sector. The main use case, motivating the work, is the prediction of cosmic radiation within the heliosphere, an investigation led by Dr. Stefano Della Torre of the INFN section of Milano.
Bando a Cascata del Centro Nazionale ICSC – “National Centre for HPC, Big Data and Quantum Computing” – codice CN00000013, Spoke 3 – “Astrophysics and Cosmos Observation” a valere sui fondi PNRR assegnati al Programma HPC finanziato sui fondi PNRR MUR – M4C2 – Investimento 1.4 con Decreto Direttoriale n. 1031 del 17/06/2022 – CUP C53C22000350006, pubblicato dall’INAF – Istituto Nazionale di Astrofisica, approvato con Determina del Direttore Generale di INAF del 28 febbraio 2024, numero 31.
Publications
- Della Torre et al., Validation of COSMICA code for massive stochastic simulation of cosmic rays propagation in the heliosphere, Astronomy and Computing, 55:101089, 2026
- Bacciu et al., Massive stochastic simulation of cosmic rays propagation in the heliosphere: The COSMICA code, Astronomy and Computing, 55:101043, 2025
- Della Torre et al., COSMICA: a GPU-optimized code for solar modulation studies, Proceedings of Science, 2025
- Della Torre et al., Advantages of GPU-accelerated approach for solving the Parker equation in the heliosphere, Proceedings of Science, 2024