December 15, 2014
A vehicle designed without CFD (computational fluid dynamics), crash, FEA (finite element analysis) and other simulations might be one you do not want to drive. But what many outside the auto industry don’t know is that vehicle engineers in most design areas use simulation for just a few operating conditions. Technological progress in both simulation software and HPC (high-performance computing) hardware is changing all that. Today’s Check it Out link takes you to a white paper that explores how automotive companies and suppliers can leverage HPC and CFD simulation software to optimize designs and enhance product performance, quality, reliability and competitiveness.
The key words are “and suppliers.” Many of you will recognize the backdrop for this paper even if your work has nothing to do with vehicle design: Vehicle manufacturers, responding to increased regulations worldwide and rapidly evolving customer tastes, want their suppliers to apply more engineering simulation to meet these requirements. What that boils down to is that all supply chain partners are expected to deliver parts and subassemblies optimized for performance and reliability within the context of the integrated system.
“ANSYS Fluent with PRIMERGY HPC Cluster: Ensuring Product Integrity for the Automotive Supply Chain” looks at how HPC clusters tuned for simulation and simulation software tuned for HPC can be a key enabler for vehicle suppliers. The broad outline here is that ANSYS Fluent 15.0 CFD software pre-installed on sector-designed Fujitsu PRIMERGY HPC clusters can empower engineers to study how their products interact within a complete system without expensive physical test beds. Furthermore, the accuracy of the results enables suppliers to assure upstream integrators of their design’s behavior under various operating conditions.
What makes this paper interesting are two distinct elements: a CFD analysis and benchmarks of several HPC cluster configurations. Essentially, ANSYS and Fujitsu studied the characteristics of exhaust simulation workloads to define a set of HPC clusters optimized for the job. The paper reports the findings.
The test simulation studied the transient behavior of exhaust gases within an assembly. Optimization techniques applied variations on several dimensions concurrently, including mass flow over various operating conditions, geometric adaptations to the catalytic converter inlet to make flow homogeneous and catalytic converter resistivity. The mesh type was hex/polyhedra. There were more than 7.63 million cells, over 43.7 million faces and nearly 33.4 million nodes. The physics used transient simulations with explicit time stepping for engine startup cycle.
Fujitsu Server PRIMERGY CX250 server nodes equipped with dual Intel Xeon CPU processors served as the HPC platform. Test nodes deployed members of the E5-26XX processor family with frequencies ranging from 2.2 GHz to 3.0 GHz, core counts of 10 or 12 per processor, and InfiniBand or Gigabit Ethernet interconnects. Intel MPI libraries implemented parallelism.
Many of the results, such as the performance and efficiency of processor types, are fascinating. Here, the cumulative trend across varying core counts per job showed that 12-core processors delivered slightly lower throughput than 10-core processors. Similar results, although more nuanced, were uncovered when comparing processor speed.
The paper also addresses the elephant in the room whenever you’re discussing HPC: Do end users spend more time futzing with the needs of the HPC system than they do engineering? Fujitsu sector-designed PRIMERGY HPC clusters not only are crafted around production workloads, they have a user environment intended to provide end users immediate application usage. This browser-based interface includes built-in methods with a simplified interface to set up, submit and monitor jobs on the HPC cluster.
“ANSYS Fluent with PRIMERGY HPC Cluster: Ensuring Product Integrity for the Automotive Supply Chain” provides all sorts of metrics that should intrigue engineering managers and analysts. It should also highly interest the C-level types because the paper clearly demonstrates that a HPC cluster tuned for your analysis workload can increase productivity by shaving days and weeks off development cycles. Download your complimentary copy of this eight-page paper by hitting today’s Check it Out link.
Thanks, Pal. – Lockwood
Anthony J. Lockwood
Editor at Large, Desktop Engineering
About the Author
Anthony J. Lockwood is Digital Engineering’s founding editor. He is now retired. Contact him via [email protected].Follow DE