Computer aided engineering (CAE) is a fundamental for modern industrial design and manufacturing, in which a supercomputer is used to simulate engineering design, or the structure, performance and status of an industrial product. With rapid development of high-performance computation technology, CAE is widely applied in aerospace, automobile manufacturing, petrochemical engineering, civil construction, etc., simulation scale is larger and larger and simulation conditions are more and more similar to real physical scenarios. Under these circumstances, CAE requires more and more computational resources and, under normal conditions, only nodes with high floating point computation power, and low-latency computational network can meet the requirement from CAE.
The fluid dynamics software FLUENT is one of the most popular CAE software applications. With rich physical models, advanced computational methods and powerful pre and post-processing functions, it is widely applied in aerospace, automobile manufacturing, environmental engineering, etc. Inspur has conducted systematic research on FLUENT for many years. We use fluid dynamics simulation of a certain vehicle as an example for a network of 14 million unstructured grids, finite volume method and standard κ-ε turbulence model. Using TEYE for monitoring of the application running features, indicated that the CPU utilization could run nearly 100% busy if we had continuous data streams communication over a network relatively sensitive to latency. Therefore, FLUENT is computation intensive and network intensive.
The characteristics of the above-mentioned typical application show that most of the CAE software programs are computation intensive and a CPU with high floating point computation capacity can help to improve the performance. Additionally, most of the CAE software programs are network intensive and a high-bandwidth and low-latency computation network should be used. Such fluid dynamics applications as FLUENT are suitable for running on a multi-node and high-performance computer cluster due to their high scalability. Structural applications, such as ABAQUS, etc. are more suitable for running on a multi-processor node due to their low scaling. In addition, structural applications have certain requirements for memory capacity and memory bandwidth.