20 Apr 2017 Stuttgart - Holm Landrock explains the NEC Aurora vector supercomputer board in Data Centre Insider. It is a kind of "GPU" but then originally designed as HPC system, not as graphic processor. The performance is 300 Glop/s per core. Interconnect throughput is at 150 Gbit/s. HLRS in Stuttgart has a NEC vector supercomputer based on NEC Aurora boards with a peak of 16 Tflop/s in production.
The article in Data Centre Insider is in German.