There is nothing so surprising about this design.  They have simply down-clocked the chips and doubled the number per board.  They could have done this in June and had the most efficient machine on the market but the chips were worth more at the time (at least that was the thinking) so it didn't make sense.  

 Here's the math:
 240MHz/270MHz(clock ratio or prisma to tube) * 800(1 Tube hash rate) * 2 (for the doubling of the chips) =1.4 TH/s

 With some extra cooling and a little know-how, you can run at 1.6+ THs with Tube-like effifciency. (NB over-clocking the proprietary design is not recommended as it could result in a damaged product)

 While I'm not impressed by the new design, I am immpressed with AMs ability to maximize the value of their product as computational difficulty rises.