I've observed that people doing big data typically have systems backgrounds while while those doing HPC are usually specialists in something else (graphics, speech recognition) who stumble on HPC because it's necessary. Of course, this is changing...machine learning people now need big data.

Companies are also forcing the two communities onto the same clusters, so much hilarity ensues when trying to satisfy both of their IO, fault tolerance, affinity, and processing needs together. Also, god forbid if the HPC people are into CUDA as well as MPI...now each machine needs a couple of pricey GPUs.