1500 Mhash (about 50 1070 or BIOS modded RX 470/480/570/580) but no power number quoted.
At 1000 watts (LOW for a Bitmain miner) it would only be 5x or so as efficient as current GPU mining.
At 1500 watts, it would be LESS THAN 3x more efficient.
Factor in the probably efficiency improvements of the next-gen Nvidia cards (due later this year SOMETIME) and the numbers get even less impressive.
Given the HUGE amount of ETH hashrate alone, even at Bitmain's typical "3000 or so" miners per batch it would take quite a few months for their ASIC to capture *HALF* of the ETH total network hashrate.
Then add in the existing total network hashrate from the other ethash coins like ETC.....
Yes, it's a threat - but it's not going to kill ETH mining soon, and unless they price it CHEAP it's likely to never make enough to pay for itself.
Unlike SIA and X11, it's NOT going to kick the hashrate multiple times higher in less than a month - it's going to be closer to the move in Scrypt where the L3+ and Innosilicon miners are ONLY NOW finally getting deployed in enough numbers to replace the "previous generation" like the A2 and the Titan and make them generally unprofitable.
IMO if Vitalik and crew keep their current target timeframe for POS deployment and actually MEET it for a change, it'll be a waste of time doing a "algo change" fork.
I don't class this miner as a "big mistake" like the Cryptonight ones from Baikal and Bitmain, but it's not looking a TON better.
https://etherscan.io/chart/hashrate shows an all-time high spike on March 26, but seems to have been a "day of the week" or "luck" thing as the trend has been a gradual DROP for about 2 weeks now.