Hi,
You are right it was 3 650 Watt PS's for only one S7, but you are wrong on the fact that the machine would not work. It's works fine and would give me an avergae of 3.848 TH based on a sample of a bit more than 5 days, however for this to happen I had to lower the freq to 579.17 the HW error rate was 0.01% and it was pulling ~ 850 Watts at the wall on 240V AC. I bought the PS from bitmain and I am now getting 4.654 TH based on a 12 day sample (each 15 minutes) with a 0.0002% HW error rate, the three boards are respectively at 46, 47, and 49 degrees Celsius and the two fans at 3720 RPM each. The one only real amazing bizarre thing I have noticed is that when I take a voltage reading with my voltmeter it sometimes "bug's" the machine were it consumes much more power and basically gives no TH's, very weird, simple reboot solves the case. My theory (with help of my cousin who is electronician) is that the test leads of the (cheap) voltmeter act as an antenna and transmits noise/interference to the controller board causing a dysfunction.