I read this thread with great interest until I got about halfway through and plugged some numbers from RichBC's original table into this calculator
https://www.cryptocompare.com/mining/calculator/btc?HashingPower=330&HashingUnit=GH%2Fs&PowerConsumption=89&CostPerkWh=0.12 and realized there's mostly no such thing as a free lunch. The difficulty has increased in the 2 years or so since this was written but still I wouldn't mind picking up an S5 cheap, before I read this I thought I couldn't afford the wattage. They aren't quite a dinosaur. Mostly free solar electricity puts efficiency in a different light. I kept wondering why Sidehack's Compacs and 2pacs were a good thing but the original S5 that the ICs came out of weren't. It's because you can play with parameters like undervolting, at least changing the core voltage.
Plotting numbers from that table it looks like things are mostly linear, there's no magic knee or obvious sweet spot.
You turn up the voltage, the watts and GH go up with it. Under 11 volts looks good though.
I kept wondering why you kept messing with commercially built power supplies when it's so simple to just start from scratch. Just some transformer so you don't have a hot chassis, a bridge rectifier, as much capacitance as practical, then start regulating it back down. If your line or load regulation aren't perfect then your GH/sec is going to be not quite what you want for a second or so, no big deal. I've built mostly linear regulators in the 1980s, but we didn't have ICs like the LM2596 and MC34063 then. You add pass transistors on the output to handle more current than the bare ICs can handle. A single heatsinked 2N3055 is good for 30 amps. I did find by experiment that the more gain stages you have the better the load regulation is. Transistor gain is current gain. The final stage might want to be 2 or 3 transistors in parallel.
Just connect them emitter to base like the last couple stages in the IC are, with some pull-down resistor because you're switching. This is from an MC34063 data sheet but I think the LM2596 will also work with external transistors. Or FETs if you prefer. You're switching, so the transistors are saturated, you aren't going to have a lot of heat, just the transistor's forward voltage drop of 0.6 volts times the current. Some junk computer PSU will probably yield you some pass transistors and an inductor. You can put a voltage adjust pot on the sense input or the IC's voltage reference, look at how an LM723 is set up. I have a bench supply I built with LM723s about 40 years ago, it's still going. Strictly linear so it's inefficient and runs hot but it's charged a lot of car batteries over the years. The diode should be a Schottky here. Keep leads short and use a toroid inductor, this is going to throw out some radio trash. Building it into a metal box wouldn't hurt.
tell me why when i hook up a buck converter outputting 12 volts and what should be plenty of amps my meter goes black and the board isnt powering up, im even running the control board and the other hashing board off a real power supply and turned it on afterwards...
