About this Abstract |
Meeting |
2025 TMS Annual Meeting & Exhibition
|
Symposium
|
Algorithms Development in Materials Science and Engineering
|
Presentation Title |
Toucan: Revolutionizing Grain Growth Simulations with Parallel-in-Time Scalability |
Author(s) |
Benjamin Stump, Samuel Reeve, Matthew Rolchigo, Daniel Arndt |
On-Site Speaker (Planned) |
Benjamin Stump |
Abstract Scope |
We present Toucan, a performance-portable code for predicting the solidification microstructure evolution in additive manufacturing, which leverages the Kokkos library and the newly developed DECA algorithm (Discrete Event inspired Cellular Automata) for time-parallel simulations of grain growth. This time-parallel approach allows Toucan to scale across multiple GPUs, essential for large-scale solidification microstructure simulations of additive manufacturing (AM) processes on modern high performance computing (HPC) systems. By utilizing features of the Kokkos performance portability library, Toucan accelerates the solution of solidification microstructure simulations by 100x compared to current state-of-the-art cellular automata (CA) codes on a single GPU. When deployed on 8 nodes of Frontier, the world’s fastest supercomputer, Toucan achieves a speed up of over 4000x over existing CA codes. As more computational resources are allocated for larger simulations, the performance advantage of Toucan continues to grow, making the possibility of part-scale microstructure simulations in AM feasible for the first time. This presentation will cover the DECA algorithm, the development and implementation of Toucan, and demonstrate performance benchmarks. |
Proceedings Inclusion? |
Planned: |
Keywords |
Solidification, Computational Materials Science & Engineering, Additive Manufacturing |