What might be a consideration is that the design is robust and although rated for less on the 12v line it can provide more anyway.  

If so you'd be moving the switching supply design very far away from switching supplies usual comfort zone of 50% full load.  Expensive switching supplies can maintain efficiency across much of its range.  

Still, switching supplies, even out of their 50% comfort zone, are far ahead more efficient than non-switching power supplies.

Wasn't saying that correctly.  A 7812 will take an input voltage of at least 15 volts.  To get the 15vdc you want to take peak, (15v)(1.414)=21.21v; then for effective, no dropouts, take .86 of that (21.21v)(.86)=18.24vdc.That 18.24vdc will be after a bridge rectifier which is going to drop well over a volt under heavy current.  Then you want a step down transformer that will drop the 120vac down to 20vac.  At the regulator say it's 30 amps, 18v-12v=6v; (30a)(6v)=180watts dropped at the regulator.  Then the rectifier, lets say that's dropping 1½v, will be 45 watts.  Then there's the transformer inefficiency but even without its losses you're dropping 225watts for 12vdc@30amps.

So if our C1 is using 818 watts at the wall using an 88% efficient switching supply, our C1 is using 720 watts which is 60 amps at 12vdc.  Our non-switching supply example above would be dropping 450watts at 60 amps before transformer losses or 1170watts.

Okay I was looking at this as it seems to be similar:

Abstract from google listing: COMPAQ PS4060 750W. HOT SWAP POWER SUPPLY. + 385V. 660uF. 100-240Vac controller. Primary board. 3.3V/ 40A. 12V/ 25A. ML4824-1. PFC / PWM ..

http://members.home.nl/a.k.bouwknegt/index_bestanden/Diagrams/PS4060%20single%20line.pdf
-------------------------

Are you sure the single 12v rail is suppose to supply 62.4A?  Are there other voltages?

Pull the ribbon cables from the controller board and measure the resistance in both directions on each PCIE connection.  If the one that burned the connector shows different, it's suspect.

So, the z750 supplies work under heavy load.  Typically a computer supply will claim a certain wattage capability and on the side list the current available on each voltage but don't supply that total wattage to any one voltage.

Worth using....  I'm retired living in the deep south, an emigre from Long Island's north shore.  I have no income aside from social security and mining.

So,  my calculation in the earlier post shows when income from my least efficient miner goes negative at current btc prices.  Easy enough to calculate when low btc value makes the miner go negative if the difficulty doesn't change much.  As long as the heat is getting exhausted and not loading down my AC there's no reason not to run my least efficient miner.  There's no reason to run it when it's take goes negative relative to power cost.

If I had the financial leeway to put the mined btc aside until payout halving next year, when I expect the value of btc to increase, I would.  But the power cost is such that I need to cash my mined btc in.  That negates any expectation of btc value increase for me on my mined btc.

Small profit margins, yes.  Just did a calculation:  average total TH/s speed of my miners according the pool, divided by the hashrate of my least productive miner, that quotient times that miner's wattage extropolated to a month, times the cost of power.  This would give me the cost for that miner to earn the monthly take, that derived from last payout divided by number of days times 30½.  If my month's btc were earned by miners running at the rate of my most inefficient, would it be more than the cost of electric?  So far yes, but I can't splurge on unnecessary power supplies or that goes away pretty quickly.

Yes.  Hadn't seen those.  I've been using these 12vdc 360watt supplies since my Mercury arrived.  

Okay, just bought one.  Looking at the price of breakout boards I think I'll go in and hard wire my cables for an initial test on an S3 to compare efficiency with my supplies.

What advantages do the reduced functionality breakout boards from GekkoScience have may I ask?

Also, is the z750p the same as a Compaq PS4060?  The schematic for the PS4060 shows rated loads for each voltage which add up to something between 600 & 700 watts and then there's the supply power load itself which would bring it up above 700.  So the PS4060 supply is rated for 25 amps on the 12vdc line or 300 watts.  So, it has less power where I want it than the 360 watt 12vdc power supply and isn't adjustable?

So, if these are good for 25 amps on the 12v line, and I use two of the 360w, 30amp supplies for an S3, then with the bboard and cables you say amounts to 50-60 bucks, I'd be spending 100-120 to do what I can do for $50 plus cables.  I don't NEED to add the filter and probably don't NEED to add the bridge recitfier heatsink nor the 20 turn trip pot instead of the stock 1 turn.  So, I probably made a mistake buying the Dell Poweredge 750w supply.

Okay.  So, $0.0779 will be your power cost.  Down here where it will get hot this summer my bill shows power to cost $0.075 but when I add in taxes but  not line charges I see $0.0805 but it's not all that clear with their adjustments.  

I worry about the heat damaging the ASICs.

To be honest, using the Kill-A-Watt, I don't see a big deal in cost savings when tuning for lower energy but I haven't tried the big jumps down.

Interestingly, with your numbers for C1 run at almost stock frequency and voltage, your 1008 / 802Watts @ .74v, compares well with these I took: after 1 hour at 250M/0760 I had 1008.57 / 818 watts for .811w/GH/s but the same unit at 2 hours showed 1002.88 / 819 watts for .816w/GH/s.  Ambient was 80°.  So reading of 802 watts is 98% of my 818watts.  That's only a 2% difference in supply efficiency.  

I'm using four 360 watt LED power supplies each modified with a heatsink added to the bridge rectifier, the one turn pot changed out for a Bourns 3299w-501 500 ohm trim pot that has, what, 20 turns from 0-500 ohms for more precise adjustment of the output voltage, and each supply has added a 6 amp 120vac line filter (reducing noise so it doesn't disturb my TV watching/recording), but, each supply costs under $25 with free shipping.  Each supply drives one blade.  I don't put them in parallel anymore as it's easy to burn one up in parallel.  Then again I had to build my PCIE harnesses.  Also I add 3½ digit LED voltage readouts to each as it's easier to tell if there's a problem.

Okay, in the post he says the most efficient test was 575-590watts, 804-817GH/s so the best would be (575w/817GH/s) or .703w/GH/s.  The 1008GH/s test doesn't show a wattage.  His highest frequency test 1123GH/s drew 943watts or .839 w/GH/s.

Wish he had provided the watts at stock settings but yes, these are good numbers, better than mine.

True.  I wouldn't run it without the resistor to a low voltage pin.  KnC had a problem early on with Jupiters blowing up a capacitor.  My theory is that without the actual motherboard the low voltage wasn't there for a start up current ramp and as a result some power supplies hit the miner with voltage much higher than 12vdc blowing up the cap.

Look for this on ebay "JUMP START EPS/ATX 24 PIN POWER SUPPLY WITHOUT A MOTHERBOARD MADE IN USA"

Oh,my mistake.  You mean for the jumper on a computer power supply motherboard connection to fool the power supply into acting as if it is connected to a motherboard.  I misunderstood.  I power my S3's with extension cord wire to PCIE connectors from switching supplies that have screw type barrier strips for the output.  So, an S3 is going to have 4 connections each good for 15 amps.

The numbers you seem to have faith in claim that at 67% of normal wattage he's getting only a 79% decrease in hashrate.  That's even if his wattage is correct at normal frequency and voltage.

At normal frequency and voltage he claims 790watts.  At normal frequency and voltage after an hour I had 1008GH/s but at 818 watts and after 2 hours I had 1002.88 @ 819watts.  

I see from numbers on the S5 forum that my cheap power supplies run about 8% less efficient.  If I reduce my power usage by 8% I get 753 watts.  So, maybe my supply isn't 8% inefficient.

At 250M/0760 I got 1233GH/s/w or .811w/GH/s.
At 250M/0675 I got 1228GH/s/w or .814w/GH/s
At 243.75M/0675 I got 1242GH/s/w or .8059w/GH/s

So, you can see why I find it incredible that somebody went from .783 down to .66 because we're talking about miner efficiency change not supply efficiency change as supply efficiency shouldn't change that much in that wattage span.

As I said previously, I did see better figures at 200M on an S3 at stock voltage but it was during the coolest part of the day and ambient temperature seems to be a critical factor when measuring efficiency.  

I have not tried 200M/0675 yet but I will eventually.  

Oh, and my pumps and radiator fans are on a supply divorced from the C1 and not included in the calculations.

I ws looking at the .66 which would be watts per gigahash not megahash?  The efficiency would seem very good unless you're off by the difference between megahash and gigahash which then would be quite bad.  I looked on the S3+ for the frequency and voltage which might correspond as S3's have the same chips as the C1.

But I couldn't find your reference.

Could you locate that post please?  Your's seems to be the first reference to 0.66W/Mhs on this or the S3+ forums when I search 0.66W/Mhs.

Sure, and somebody takes an old powered amplified speaker pair from a PC, cuts off the speakers and tries running an S3 though that.

I've seen speaker wire that's a good 16 gauge but you'll find lesser stuff too.  Can't go wrong with 16 gauge 15 amp extension cord wire with ends cut off as long as you don't confuse polarity.  I've seen speaker wire with clear insulation have one conductor stranded copper and the other some silver colored stranded so there's that but extension cords have one conductor ribbed.  The ribbed conductor is always neutral in a 120vac application but I always make that the red and the non-ribbed black, kind of revers of the AC application.

Characterizing my S5 @ 350M and see there's no voltage adjustment and ran a search for asic-freq but found nothing.  I realize the voltages per frequency wouldn't be applicable to S3+'s as they're different ASICs but was curious.

The term 'speaker wire' is kind of vague.  Speaker wire can vary in gauge.

Just took the top off the 'new' bottle of liquid flux.  The foil seal is broken, there's flux removed, and there's crap around the inside top.  Bought from eio-electronicinventoryonline on ebay.
---------------------
This size smt is quite intimidating.  After reflowing the solder at the PCIE connectors I reflowed solder at the components top inside of the worse of the two hashboards.  This size is so difficult that I'm only going to reflow at the PCIE connectors on the other board, reassemble and test.  If something went bad from my reflowing then I want to know.  It seems a high probability that there may be current choking at one or more pins of the PCIE dropping the voltage so that's what I'll address on the second board.

-----------------------

Back up and running.  All ASICs are o's.   Both fans running.  Will see how it does on the hour for a few.

----------------------

I restarted both S3#2 and S3#1 at the same time to compare - although S3#1 started off all warmed up.  At 30 minutes it looks like my work on S3#2 was no help as it's down to 437.00 avg and dropping while S3#1 is 442.89.

Ran 45 minutes then hashrate had dropped to 420 and as it was on the half hour when I have cgminer-avg-monitor run, it restarted cgminer.

------------------------

Lesson learned: Only buy Bitcoin miners from the manufacturers except BFL.  One might luck out like I did with Cryptocrane but more likely you'll get somebody's castoff like from Pines in Florida.

My flux arrived so tomorrow while fresh I'll rip down my lousy S3#2, removing all four heatsinks, cleaning well, then hitting each solder point lightly to reflow the solder, well most of them anyway.  Won't touch the ASICs.  Reclean, reassemble and Bob's your uncle, hopefully.

Return and repair takes some time.  I think I'd first check continuity between the ribbon cable pins and destinations.  Perhaps not all points can be seen.  Is it always the same side?  If you put a single board to the left (looking at the RJ45 port) does chain# 1 show working ASICs?  When you connect 2 boards which chain works? When you have a single board connected chain#1 is on the left.  When you have two boards connected chain#1 is on the right and chain #2 on the left.

Isolate specifically which ribbon cable connector to investigate first.

Also you can take the controller board out into bright sunshine and go over it with a dental pick looking for metal or solder balls that don't belong.