�ݺ�ߣ

Editor's Notes

• #7: This slide demonstrates the improved scaling of Distributed ANSYS in R18.2 compared to R18.0. This is a ball grid array (BGA) benchmark from MicroConsult in Germany involving a nonlinear transient analysis with creep material properties (no contact). The sparse solver is used to solve the 4 million equations. There is 1 load step with 12 substeps and a total of 25 iterations to reach convergence in the nonlinear analysis. This model was solved on the Intel Endeavor cluster. Each compute node contains 2 Intel Xeon Gold 6148 (Skylake) processors, 192 GB of RAM, local SSD storage, and RHEL7.3. These runs were performed with the Intel Omnipath interconnect. Here we are timing the “Elapsed time spent computing solution”. So this measures the entire solution (forming elements, solving the equations, computing stresses/strains, etc..), but does not include items such as resuming the database or combining files at the end of the solution. One can easily see that with R18.0 the scaling stops at approximately 1000 cores. However, with R18.2 the scaling continues to at least 3072 cores.