Skip to main content

MPP Scalability

MPP (short for Mas­sive­ly Par­al­lel Pro­cess­ing) is a type of com­put­ing avail­able for LS-DY­NA that us­es many sep­a­rate CPUs run­ning in par­al­lel each with their own mem­o­ry to ex­e­cute a sin­gle analy­sis. In or­der to solve large im­plic­it CFD an­a­lyzes, it is im­por­tant to pro­vide a good CPU scal­a­bil­i­ty in or­der to ac­cel­er­ate the analy­sis and save some com­pu­ta­tion­al time. A nu­mer­i­cal mod­el of around 2 mil­lion de­gree of free­dom was run as a stand alone CFD mod­el. An FSI analy­sis was al­so per­formed on the same mod­el bring­ing the to­tal num­ber of de­grees of free­dom to around 3.5 mil­lions. Bear­ing in mind that the ICFD solver is im­plic­it, the re­sults show good speedup ca­pa­bil­i­ties : of 40 for 128 cpus in the CFD on­ly case and a speedup of 55 for 128 cpus in the case of FSI. For the next de­vel­op­ment cy­cle fur­ther im­prov­ments will be im­ple­ment­ed.