It would be odd for a fan to draw more power than expected, although not impossible if the bearings are wearing out.

You might want to consider buying a replacement fan for ~$15 USD and see if the problem goes away.

Agreed.  How far is your genset from your devices?

Please also confirm your talking about a motor/generator device - which does do a good job of power isolation.

Curious why you are using one though???

Please provide a photo of your setup so we can visualize what your doing. 

This is a classic case of a photo being worth a thousand words.

I'll 2nd this comment.  My day 1 experience when I turned on (80) APW3++s at near 0C was that less than a third would boot an S9.  At 10C they mostly worked (I found a couple of them that will only power 2 cards).

To this day, if I suspect an APW3++, I take it into my hot aisle for 30 minutes or so and then try it again.

Once they power up, they seem to generate enough internal heat to keep going.

If you can not direct vent the fans into some form of hot aisle, I would highly suggest you follow the above advise and use 6".  You could alway make adapters for the fans out of stryro or double thickness drywall.

Quick math, presuming some comfortable margin (7 amps per miner).  At 80% breaker load = 8.75 breaker rating amps per miner.    Anyhow, presuming each leg provides 150 amps maximum (limited by a breaker), you could have 150/8.75 = 17 miners per leg.  Roughly.  Thus 51 total.

Do you really have 240V 3 phase?  208V is much more common here in the USA.

And what does the kernel log show?  Sounds like one of the fans isn't coming up (although it would spin from the force of the other one).

You sure about that?  I was under the impression a X amp 208V 3-phase supply could provide X amps to each of the A-B, B-C, and A-C legs - presuming those were separated out as they normally are for data centers.  Balancing the load between the legs is the only tricky part to optimize utilization.

That said, I've not worked with 3-phase, so don't know for sure.

Please tell us more.  I would think a motor/generator (what I think of as a genset) would produce fairly constant power, since minor variations of a few cycles would be leveled out by the inertia of the unit.  That said, its ALWAYS a good idea to protect electronic equipment with surge protectors.  I have one on each of my mining distribution panels.  Presuming your miners are close to your distribution panels, this is very cost effective.  Believe I paid something like $45 each from my electric Coop for mine and each one protects 20 miners.

You may be low on the electrician.  I had one quote me $10K to hang (5) panels and hook up to the transformer.  That did not include any breaker work inside the building beyond the panels.  Are you going to do all your own branch work?  Will your city permit that?

You can't do much about the noise due to the ventilation requirements.  I presume you have factored in something to deal with the ventilation problem?

---

(Moderator's note: This post was edited by frodocooper to remove an off-topic portion of the post.)

You can try SSHing in with root/admin...

Same link I provided in my initial response. 

Any relatives with real-life farms that like you?  Perhaps you could use an old out-building that is near power.

Most working farms, like mine, have near zero regulations/permits/inspections/etc.  That allows me to do my own work, which really keeps the cost down.  Of course, do everything by code - those were created for a reason.  Oh, and "do my own work" includes things like pulling meters and wiring up primary electrical panels to the transformers -  with the blessing of my local power coop.  Of course, if your not REALLY comfortable doing that type of stuff, hire a licensed contractor.

Presuming you use natural gas for heat, hot water, and a dryer, avoid using AC in the summer, and have a fairly small house.  Most newer USA homes have 200 amp service for a reason, the oh-so-popular mini-mansions have 400 amp.

It also presumes your allowed to do light-commercial work in your neighborhood, which most won't allow.

Have you already ordered your miners?

There are a LOT of post explaining this, but one more time, in round numbers:

Frequency * power consumption / joules per GH = chip power consumption
chip power consumption / power supply efficiency = 12V power drain, shy a bit for fans, the controller, etc.
wattage / voltage = amps (remember Ohm's law???)
breakers should never be continuously loaded for more than 80% of their rated capacity
wire should always be rated at the breaker rating (or larger, but nobody commonly does that)

so:

13500 MH* 0.098 J/MH = 1323 watts (just like the Bitmain page says)
1323 watts / 0.93 efficiency = 1422 watts, call it 1450 with fans an controller for this exercise
1450 watts / 220 volts = 6.6 amps
20 amp breaker allows you to use 12 gauge wire, which is easier than 10 gauge if you opt for 30 amp breakers
20 amps * 0.80 = 16 amps allowable per breaker
Therefore:  You can put 2 miners on a 20 amp 220V breaker

(75) S9s would require (38) 20 amp 220V breakers.

A 200 amp breaker panel has a 200 amp master breaker at top.  80% rule applies to that too, so 160 amps total usable per panel.

160 amps / 6.6 amps (which is pushing it to the limit) = 24 miners per panel or (12) 20 amp circuits, but...

A typical 200 amp indoor main breaker panel has 30 spaces, 60 circuits, 15 spaces per side.  Each 220V breaker requires 2 spaces, so you can only get (7) 220V breakers in per side, but that would overload the panel since 7 per side = 14 breakers = 28 miners = 185 amps which is greater than the limit.

You can find 20 space, 40 circuit 200 amp panels that work out better:  20 spaces = 10 breakers = 5 per side.  10 breakers = 20 miners = 132 amps which is comfortably below the 160 amp limit.

Using those 20 space panels would mean you would need (4) such panels.  To power them, you would need 75 * 6.6 = 495 amps of continuous draw power spread between those panels.

That is excluding power for lights, fans, etc. (not to mention the rest of your house)

Then there is the minor problem of getting rid of 1450*75 = 108,750 watts of waste heat, but that is a different discussion.

Oh, and dealing with the noise.

Moderator:  if you are willing, could you Pin this?

My recommendation:  Swap the hashboards (or hash cards) and isolate the problem.

If your at 2 months, your still under warranty.  If you can prove its a controller, Bitmain will send you a new one.

Please keep us posted on how long your free electricity remains free.  I suspect somebody is going to eventually notice the huge jump and initiate a conversation with you about it.

So 3%.  Been awhile since I checked out household goods, but thought riders for those were around 2%.

BTW - how did you value your S9s and T9s?  Purchase price?  Replacement price?  E-bay price?  If replacement, how did you figure in the lot limit that Bitmain has recently been imposing?

Miners consume power based on their technology (typically measured in "nm" (nanometer transistor size) and the frequency they switch those transistors.  Smaller transistors use less power than larger ones, which is why 16nm S9s are more efficient than 28nm S5 technology.  Of course, design changes within a tech level affect power usage as well, which is why S9s are better than S7s.

This is why the faster you run an S9, the more power it requires:  1127W (11.5TH/s batch), 1225W (12.5TH/s batch), 1274W (13TH/s batch), 1323W (13.5TH/s batch), 1372W (14TH/s batch)

Frequency itself is limited by the power requirements and the corresponding heat generated (remember that every watt of power used switching results in a watt of heat being generated), and by the leakage rate of the transistors themselves.   That leakage is why in any given batch of chips, some will run faster than others - over simplified, the leakage is basically caused by small errors in the printing of the transistors.  Better/faster chips have less random errors.  Think of it a bit like ink bleed, or a small hole in a bucket.  If you dump the bucket slowly, you will get a lot of water out of it.  If you dump it faster, the leak caused by the hole becomes more significant.  Thats not a perfect analogy, but hopefully helps.

Bottom line:  No.  Even if you had a 10KW power supply on a 14TH/s S9, it would still only draw about 1.4KW.

Watts are watts.  A 50W light bulb pulls 50Ws if run at 12V (~4.2 amps), 110V (~0.45 amps), or 220V (~0.23 amps).  Because of the lower amperage draw at higher voltages, thinner wire can be used to carry the current.

e.g.  If your running a 1350 watt device, you need at LEAST a 1350 watt power supply/power supply efficiency.  For example:  if your power supply is 80% efficient (cheap and bad!), it would need to have 12V rails rated at 1350/.80 = 1688 watts and would be running at 100% utilization then, wasting (and generating heat for) 338 watts of power.  At 93% efficiency (APW3++), that math works out to 1452 watts with a waste of only about 102 watts.  You also need to check the efficiency curve for the power supply, since "efficiency" is often quoted at the optimal power draw for that device.

From the Bitmain page:  Power Consumption   1127W (11.5TH/s batch), 1225W (12.5TH/s batch), 1274W (13TH/s batch), 1323W (13.5TH/s batch), 1372W (14TH/s batch) (at the wall, with Bitmain’s APW3 PSU,93% efficiency, 25°C ambient temp). Expected discrepancy of +10%

The above statement from BitMain is either miss-leading, or lost something in translation.  At 0.098J/GH, a 13.5TH machine would consume 1323W JUST FOR THE CHIPS.  Their comment to add 10% however should compensate for the fans, controllers, support logic, and power supply inefficiency if your running one whose sweet spot is near 1350 watts at 93% efficiency, like the APW3++ is.

Regarding 3 Phase power, unless you have a 3 Phase power supply (VERY unusual), that 3 Phase power is likely just going to be broken into 3 single phase distributions.  Your challenge is to keep the load balanced across those distributions.