Yep there's 'a bug' as he says:
https://www.youtube.com/watch?v=nLRCK7RfbUg&t=9m20s
AdoredTV is right. Nvidia needs to sort that out before Vega. Their driver can't feed their card fast enough, whereas the AMD driver can..
EDIT: I wonder if this is due to the GPU architecture, meaning that async compute stuff really provides for that huge an impact. If that's the case and we're seeing the true limit of the Pascal architecture (i.e. it's a hardware limitation and that's the best their software can do), then it would be really bad for NVidia... Having said that, the difference seems to huge for that to be true so maybe it's just the NVidia software (which CAN be fixed).