1500/34000 = too good too be true.
44 watts a th is head and shoulders better then the m10 which is about 68 to 70 watts a th.
But they did claim under 50 watts a th.
Same ballpark as Bitfury's "Clarke" chip.
Bitmain does have a history of getting higher performance out of a node than most of the competition (though Bitfury is usually in the same ballpark).
Innosilicon in particular tends to be lower performance on a given node than Bitmain manages (there are persistant rumors that the Innosilicon A3 was never released because it was no better to worse performance than the chip that Bitmain used in the S7 on an older node).