Performance analysis and optimization of G+Smo and IgANets on LUMI (EINF-12250)
The aim of this project is to evaluate and optimized the computational performance and parallel efficiency of G+Smo and IgANets on the AMD-based supercomputer LUMI.
Duration
24/12/2024 - 23/12/2025
Collaborators
Henk Dreuning
IgANets − Physics-informed isogeometric analysis neural network (SURF-ETP0001)
The aim of this project is to port all custom compute kernels of our IgANets software from CUDA to HIP/ROCm and perform a performance evaluation on AMD MI210 graphics cards.
Check out our summary of results.
Duration
13/10/2023 - present
Collaborators
Xavier Alvarez Farre
Promising Application: Computational steering: Interactive Design-through-Analysis for Simulation Sciences (EINF-7907)
The aim of this project is to develop a collaborative and interactive design-through-analysis workflow atop the G+Smo isogeometric analysis library and our recently developed IgANets, a neural network approach towards isogeometric analysis.
Check out our online demonstrator.
Duration
16/02/2023 - 15/08/2025
Collaborators
Casper van Leeuwen, Paul Melis
Promising Application: High-performance Isogeometric Analysis for Dynamics (EINF-3689)
Many engineering problems are modelled by time-dependent partial differential equations that are nowadays solved by sophisticated numerical simulation tools. While parallelization in space is common practice nowadays the efficient utilization of parallelism in time is not so standard. The aim of this project is to benchmark and if needed tune our Multigrid-in-Time Isogeometric Analysis code framework (G+Smo + XBraid) which has been tested successfully on Lisa for up to 2048 cores at large scale.
Duration
16/08/2022 - 15/08/2023
Collaborators
Benjamin Czaja