Stella.jl: Fast, Open-Source, Modern CPU-GPU DEM Solver
Please login to view abstract download link
The Discrete Element Method (DEM) is essential for simulating granular materials, yet the current software ecosystem is fragmented. Researchers and engineers often develop separate solvers, resulting in duplicated efforts, inefficient development, and limited collaboration. In response, we introduce a unified, open-source DEM framework built in Julia that combines user-friendly design with high-performance capabilities. The framework supports CPUs, all major GPU architectures, and distributed computing—scaling seamlessly from personal workstations to high-performance clusters. Benchmarking shows a performance gain of over 2× compared to LIGGGHTS—one of the fastest open-source CPU solvers [1]—alongside an intuitive interface and streamlined simulation setup. Designed from the ground up for modularity, the core code base facilitates the straightforward addition of new contact models and coupling strategies, while advanced BVH contact detection, and state-of-the-art numerical integrators maintain exceptional speed and efficiency. By consolidating efforts into a single high-performance package, we hope to reduce redundancy, foster collaboration, and accelerate advances in particle simulation research for all. References [1] Dosta, M., Andre, D., Angelidakis, V., Caulk, R. A., Celigueta, M. A., Chareyre, B., Dietiker, J. F., Girardot, J., Govender, N., Hubert, C., Kobyłka, R., Moura, A. F., Skorych, V., Weatherley, D. K., & Weinhart, T. (2024). Comparing open-source DEM frameworks for simulations of common bulk processes. Computer Physics Communications, 296, 109066. https://doi.org/10.1016/J.CPC.2023.109066