Using a Rolls-Royce representative engine model to evaluate scalability of LS-DYNA thermal solvers

In the Finite Element Modeling community there is a trend to use models with increasing modeling details which raises the numbers of elements and solution variables. The increase in solution variables has a big impact on the run time of the analysis. Reducing wall clock time is an important item in using numerical analysis in production. The wall clock time can be reduced by using improved CPU technology and hardware with a higher throughput and lower latency for memory, storage and interconnect. On the software side, the use of parallel models to utilize more cores in an analysis reduces the wall clock time. Key measure for reducing wall clock time is scalability, which is in general expressed as the reduction of the run time due to an increase of cores used for the analysis. LSTC is currently offering LS-DYNA in three different parallel models, namely shared memory parallel (SMP), massive parallel processor (MPP) and the combination of both models (HYBRID). The focus on these developments is scalability for all three parallel models. Scalability is influenced by several factors. Beside the already mentioned hardware environment, main contributors are the decomposition (MPP and HYBRID) of the model, the model size and application type. Scalability can not only be evaluated on a global implementation level. It needs to be evaluated on the application at hand and the features utilized in this analysis. This contribution discusses the scalability of thermal solvers offered by LS-DYNA MPP using a surrogate engine model from Rolls-Royce. Three thermal solver types are used with three different MPP rank count (4, 8 and 16). The scalability is measured using the wall clock time summary of the LS-DYNA runs found in the d3hsp files.