CAE Heads for Extreme Scaling
Over at the Cray blog, Greg Clifford, Cray’s Manufacturing Segment Manager, writes about the evolution and current state of extreme scaling in CAE applications. Clifford claims that despite being steeped in the fast-moving field of HPC, some things remain the same for computer-aided engineering (CAE) workloads: 1) the demand for compute power has nearly doubled every year; and 2) the names of the ISV applications used for CAE simulations – e.g., NASTRAN, Abaqus, Fluent, LS-DYNA – haven’t changed much over the last two decades.
One major development that has occurred is an increased reliance on parallel processing to boost the performance of CAE apps. Clifford highlights the trend toward extreme scalability in the CAE applications, which he loosely defines as using more than 1,000 cores. He believes this is a turning point for CAE simulation.
For decades, the CAE field has relied on Moore’s Law-driven processor improvements along with increases in processor frequency to satisfy computing demand. As processor frequency has plateaued, the HPC industry has derived its FLOPs-per-processor improvements from multicore processors.
“Today it is common for large commercial CAE environments to have 10,000 cores available and several organizations are over 50,000 cores,” notes Clifford. “Of course, to leverage the increased performance of a multi-core processor implies the analysis is using more compute cores per simulation (i.e. more parallel scaling). However, most manufacturing organizations have used the increase in total compute power to increase the overall throughput (i.e. capacity computing) and have not scaled up the performance for individual simulations (i.e. capability computing).”
In this ultra-competitive era, manufacturers are dealing with the dual requirements of higher simulation fidelity and tighter design schedules. They are looking to significant boost the turnaround time on large jobs – not 10 percent faster but 10 times faster, notes Clifford. As most CAE simulations are operating in 250-core territory, extreme parallel scaling offers a path forward.
In order for this extreme scaling to really take off in CAE, it will need a high-value simulation field to take a leadership role in leveraging the technology, states Clifford. He identifies computational fluid dynamics (CFD) as a promising candidate. It will also require collaboration between the system experts and the ISVs. Cray, for example, worked with ANSYS to improve Fluent scaling in recent versions to over 10,000 cores. The figure below illustrates the performance for an automotive external aerodynamics simulation, which employed strong scaling to over 12,000 cores.