New Features in LS-DYNA HYBRID Version

Numerical noise arising from different MPP core counts compels users to fix the number of cores used by LS- DYNA® MPP e.g. during a vehicle development program. This fixed core count limits job turn-around time and flexibility in managing computing resources. In addition, using a large number of cores for calculations diminishes scalability with pure MPP. LS-DYNA® HYBRID addresses these issues through the use of both MPI + OpenMP technology. LS-DYNA HYBRID is able to produce consistent numerical results when changing the number of OpenMP threads thereby reducing job turnaround time. In addition, LS-DYNA HYBRID can greatly reduce the number of processors involved in message passing and achieve much better scalability over large number of cores. Furthermore, for the implicit applications LS-DYNA HYBRID not only reduces the memory requirement per node but also decreases IO activity. Currently, LSTC and Intel® teams are working together with a customer to evaluate LS-DYNA HYBRID code using a custom QA (Quality Assurance) suite. The consistency and performance will be discussed in this paper.