White Papers

Altair OptiStruct Performance
20 Dell EMC Ready Solution for HPC Digital Manufacturing Altair Performance
6 Altair OptiStruct Performance
Altair OptiStruct is a multi-physics Finite Element Analysis (FEA) solver commonly used in multiple
engineering disciplines. Based on finite-element and multi-body dynamics technology, and through advanced
analysis and optimization algorithms, OptiStruct helps designers and engineers rapidly develop innovative,
lightweight and structurally efficient designs. It accurately handles nonlinearity of materials, geometries, and
contact for applications including gasket analysis, bolt pre-tensioning, rotor dynamics, and thermo-structural
analysis.
Two types of solvers are available with OptiStruct: Distributed Memory Parallel (DMP) and Shared Memory
Parallel (SMP). In general, the DMP solver offers equivalent or better performance than the SMP solver
particularly when all of the cores on a processor are used. Depending on the specific problem types, FEA
simulations may or may not scale well across multiple processor cores and servers. Implicit FEA problems
often place large demands on the memory and disk I/O sub-systems. Given the varying system requirements
for different types of FEA problems, benchmarking for OptiStruct was performed using a variety of server
configurations.
For system testing, two benchmark problems provided by Altair (Engine=5MDOF, Taurus=25MDOF_FRF).
provided by Altair, are shown. Benchmark results from various systems using both the SMP solver and the
DMP solver where possible are show below in Figure 10 below.
These results are shown relative to the performance of the 13G 8-core Intel Xeon E5-2667v4 based server,
where higher indicates better overall performance. The E5-2667v4 based server had 512GB of memory with
2TB of local storage, the E5-2697Av4 server had 128GB of memory and 1TB of local storage, the 6136-
based server had 384GB of memory and 2TB of disk space, and the remaining servers had 192GB of
memory with 1TB of local storage.
The Engine model was small, making it was possible to carry out benchmarks on all test systems. The
Taurus model was much larger and difficult to run on systems with modest memory and storage amounts,
resulting in several missing table entries. The results shown for the DMP solver were those obtained using the
optimal number of DMP partitions per node. The optimal number of partitions can vary based on the
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
Engine-SMP Taurus-SMP Engine-DMP Taurus-DMP
Performance Relative to E5
-2667v4
Figure 10: Single System PerformanceOptiStruct
E5-2667v4 E5-2697Av4 6136 6142 6242 6248 6252