The MSC Nastran application is almost 50 years old, but it continues to receive enhancements, and we at Cray continue to see significant performance improvements when MSC Nastran is run on the latest Cray® CS400™ cluster supercomputer systems.
MSC Nastran is one of the most widely used structural analysis applications in the Automotive and Aerospace industries. Notably, it is widely used for large, linear dynamic analysis such as vehicle vibration simulation. MSC Software recently posted the performance results for the latest version — MSC Nastran 2014. The results proved to be interesting for several reasons, but we were especially pleased to see that our results for the noise vibration and harshness (NVH) simulations were the best MSC Nastran 2014 results overall. They were one and a half to two times faster than the comparable 2013 results.
NVH is one of the most challenging simulation fields because it stresses every aspect of a system’s configuration. As automotive companies push for higher-fidelity simulations (such as acoustic analysis), NVH model sizes have grown dramatically in order to capture high-frequency response. NHV simulations are challenging to run, but they’re also very high value to automobile manufacturers, as low noise and low vibration are some of the strongest selling points. Who wants a car with a constant vibration in the steering wheel or a rattle coming from the dash board? Aggressive use of NVH simulation can identify potential issues earlier in the design phase and ultimately enhance the driver’s experience.
The MSC Nastran 2014 results provide several insights, but two items stand out:
- Model complexity continues to increase. The largest NVH model in the results is an automotive “floor pan” containing over 19 million degrees of freedom. This is five times the size of the largest NVH model from the previous benchmarks (version 2013), and it’s an indication of how NVH model sizes are growing.
- Simulation performance times continue to improve. Compared to the 2013 performance for a “car body” model, the 2014 Cray results are 1.6 times faster. From the hardware perspective this is a result of Cray configuration featuring larger memory, a faster IO configuration and improved AVX2 performance from the Intel® Xeon® v3 (Haswell) processors.
All results are posted on the MSC Software site.
Efficiently computing large NVH simulations —large eigenvalue problems— requires a robust system configuration. In general, NVH simulations do not scale well across nodes, and they perform a large amount of scratch I/O. Typically, the best configuration for NVH is a single node containing: a fast processor (to reduce the “CPU time”); large memory; and a robust I/O system (to reduce the scratch I/O time). To properly design an NVH system, we drew upon the expertise of Doug Petesch (who supports the NVH field for Cray and brings years of experience with NASTRAN and high-end NVH). The result was a node configured with:
Cray CS400 system dual socket node
Processors: Two Xeon E5-2667-v3 (Haswell, 8 core, 3.2 GHz)
Memory: 512 GB memory
Disk: Three PCIe SSDs
Even with the large increase in this test model size, we routinely see production NVH benchmark requests from the auto industry that are larger than the “car body” test case. It is often difficult to estimate the performance of NVH simulations or to know what hardware configuration offer the best value. However, our ability to test the performance of very large NVH simulations — thanks to the availability of larger nodes (512 GB, multiple SSDs) in our CS400 system benchmark environment — allows us to directly show the performance advantages of a properly configured Cray solution.
Even for the very large and complex established application such as MSC Nastran, there are significant performance enhancements available by tuning the latest hardware and software technology to the simulation workload. These benchmark results are yet another example of the Cray application team working directly with independent software vendors and our user community to continually improve application performance and add value for our mutual customers.