GPU-Accelerated OpenSees

Exploring CUDA-based solvers and GPU computing to accelerate OpenSees structural dynamic analyses.

Motivation

Large-scale nonlinear and elastic dynamic analyses in OpenSees spend significant time in linear system solution and assembly phases. GPU-accelerated computing offers a path to faster turnaround for parametric studies and regional workflows.

Objectives

1. Profile bottlenecks in typical OpenSees dynamic analyses
2. Implement or integrate CUDA-based sparse solvers for critical kernels
3. Validate accuracy and speedup against CPU baselines
4. Document reproducible workflows for researchers using OpenSeesPy front ends

Methods

• Finite-element assembly on CPU with GPU-resident linear algebra
• Benchmark problems: SDOF to multi-story frame and wall models
• Comparison with CPU direct/sparse solvers

Related work

Conference presentation at the 2024 NHERI Computational Symposium: Accelerating Finite-Element Structural Elastic Dynamic Analysis Using GPU Computing. Upcoming related work at the 2026 NHERI Computational Symposium.

NVIDIA certification: Fundamentals of Accelerated Computing with CUDA Python.

Status

Active development. Code and benchmarks will be released on GitHub when ready.

Collaborators

Open to collaboration with researchers in the NHERI SimCenter and PEER communities.