In this podcast posted over at bitpipe, Cray’s Greg Clifford discusses how CAE applications are mapping onto new hybrid supercomputers and how to best meet diverse resource requirements of CAE platforms.
While computer-aided engineering (CAE) applications have remained relatively constant in recent years, the HPC hardware environment they run on has advanced rapidly. For example, microprocessors have “many cores,” large SMP servers feature over 100 cores and a terabyte of memory, SSDs eliminate the rotating disk drives, and GPU accelerators offer over 10X performance gains. The twist is that these advancements can both greatly improve performance for some analysis and hurt performance in other cases. Since the application resource requirements vary widely between CAE code and even between different models, the best performance requires a hybrid environment.
You’ll need to give up your email and register at bitpipe to hear this podcast, but Greg is one of the most knowledgeable guys I know in the HPC application space, so it’s definitely worth a listen.