White Papers
41 Dell HPC System for Manufacturing—System Architecture and Application Performance
Figure 36 LSTC LS-DYNA Car2Car Scaling—Explicit BB
4.6.2 ODB-10M
The ODB-10M benchmark is a simulation of a vehicle colliding into an offset deformable barrier. This
benchmark model contains 10.6 million elements. Figure 37 shows the measured performance of the
reference system for the ODB-10M benchmark, on one to eight EBBs, by using 32 to 256 cores. Each data
point on the graph records the performance using the number of cores marked on the horizontal axis in a
parallel simulation.
Figure 38 presents the same performance data plotted relative to the “32-cores (1 Node)” result. This
makes it easy to see the scaling of the solution—the performance improvement as more cores are used
for the analysis. Problem scalability for LS-DYNA depends on the number of elements and other features
being simulated in the problem such as contact or airbags. The ODB-10M benchmark scales somewhat
better than the Car2Car benchmark, because the ODB-10M benchmark is a larger problem.
This benchmark was run using both Intel MPI and Platform MPI. The performance obtained from the two
MPI implementations was nearly identical.
0
1
2
3
4
5
6
32(1) 64(2) 128(4) 192(6) 256(8)
PerformanceRelativeto32Cores(1Node)
NumberofCores(NumberofNodes)
LSTCLS‐DYNACar2CarScaling—ExplicitBB
R8.1AVX2IntelMPI5.1.2.150 R8.1AVX2PlatformMPI9.1.0.1