Developing power supplies for Bitcoin miners is in a class of problem called "bitch on wheels". This however is not an unknown issue, it has buried a number of other companies over the years and anyone who has designed high power supplies knows the joy of watching smoke and plasma balls rise from 300 amp IGBT circuits.

I think however that people designing boards tend to be simply willfully ignorant of the issues of providing 500 amps of power at half a volt and thus they screw up the same thing over and over again. At least on a SHA256 chip each engine is pulling constant power through the hashing cycle. I wonder if a scrypt miner pulls varying levels of power depending on where it is in the computations, and that power variance fucks over any attempt to provide a constant level of power to the engine with 500 other engines on the same chip bitching and moaning as they vary their requirements from cycle to cycle.

It could explain the complete graveyard in the Scrypt space. That could also move the level of "hard" from "flaming levels of difficult" to "fuck, we have a chip and can't make it work! And here comes THE NOISE FAIRY!!!!!!!!!!!!!!"