Finite element method completely implemented for graphic processor units using parallel algorithm libraries

Abstract

A finite element code is developed in which all computational expensive steps are performed on a graphics processing unit (GPU) via the THRUST and the PARALUTION library. The code is focused on simulation of transient problems where the repeated computations per time step create the computational cost. It is applied to solve partial and ordinary differential equations as they arise in thermal-runaway simulations of automotive batteries. The speedup obtained by utilizing the GPU for every critical step is compared against the single core and the multi-threading solution which is also supported by the chosen libraries. This way a high total speedup on the GPU is achieved without the need for programming a single classical Compute Unified Device Architecture (CUDA) kernel.

Modelling and Optimisation of Laser-Structured Battery Electrodes

“…The individual simulations were performed via an in-house-developed finite element code (based on [24]), which is highly optimised for single-thread transient simulations of several thousand degrees of freedom. This yielded an average simulation time of around 6 s per conducted simulation.…”

“…It consisted of a python script collection that automates the distribution of multiple simulations to a CPU cluster and the automated evaluation of their results. Each simulation then used a scripted meshing algorithm (utilizing SALOME [27]) and an in-house-developed FEM toolbox [24] that performs the simulation of the discussed cycling protocols. Altogether, about 8400 such simulations were performed to deliver the presented results.…”