The 20% derate on 24/7 operation is a NEC requirement for ALL circuits - including your infeed from the transformer INTO the panel.

 So yes, the 53 miners figure would be closer to correct, as "400 amp service" would work out to 320 USABLE amps (it's probably safe to fudge a couple of amps in that sort of a case, it's not a "OMG I went over an amp I'm FRIED" thing).


 As far as your sig goes - I prefer the variant "We are the immortals. Join us or die.".

 9-)

Roof fans and vents might be an option - or replace some of the panels in or the entire roll-up door with a bit "louvers" setup for the intake venting.

 I set the monitoring machine to use 255.255.0.0 for a subnet mask instead of the standard 255.255.255.0
 If you're using a LINUX machine to monitor, that would be a 16 subnet mask instead of a 24.

 When I had the 192.168.10.0 subnet though, that went through a secondary machine as it DID need a different physical interface to be able to talk to the 10-Base-2 (coax) subnet.
 I handled that by having 3 net cards in the firewall/gateway/router machine, one to talk to the ISP connection, one for the 100-Base-T stuff (Gigabit was still expen$$$ive then), one for the 10-Base-2 stuff.

 On most modern high-end power supplies, that 450 watts IS 450 watts of 12v capability - then the OTHER draws are pulled out of that.

 EVGA in particular works that way on most if not all of their power supplies.

 OP did mention a Delta - which I would presume is what the PCI-E power connectors to the cards themselves are running from.

 75 watts per PCI-E bus connection - you're overloading that poor little 450 it would appear, since it's probably also running the CPU/RAM/MB chipset/HD(SSD) stuff.

 The "one miner per card reporting" is a well-known DSTM issue, and has been mentioned in that thread.

 Way low for any 1080, much less that MONSTER TDP model - unless you have the TDP tuned way low or bad settings for core/memory clocks.'
 Even my unhappy Gigabyte 1080 cards manage 430-440 running around 105 watts and +100/+100 (higher core = crashy card, higher memory = lower hashrate 'till they hit crashy card on those).

 DSTM - varies, sometimes it gives higher performance sometimes lower than EWBF - and settings on a specific card can often affect that.

 BMiner - I have not and won't bother even looking at as long as it is Windows specific.

G3930 is cheaper by a couple bucks - but has Kaby Lake compatability issues with some motherboards and operating systems.
IMO it's worth the extra couple bucks to get an older Skylake part to avoid the issues.

 You cannot connect ANY risers safely to a SATA port - the actual SATA CONNECTOR is not rated for the 75 RATED WATTS POWER DRAW on the PCI-E bus (the CONNECTOR ITSELF is only rated 54, and some cards DRAW MORE from the PCI-E bus on peaks).
 Since you already have PCI-E risers, I'd recommend getting some MOLEX to PCI-E 6-pin adapters to use on most or all of them as the EVGA 850s aren't going to have any PCI-E connectors to spare.

999 confirms seems to be excessive - is that on the Bittrex end or the Poloniex end?

Between the ASICs, the GPUs that are actively mining, and my BOINC/Moo Wrapper/Dnet farms (plus a few extra machines and the evaps etc) I'm pulling ballpark 15 KW on average.
 Bit less the last couple months after I shut one evap down and the other has been getting turned off part of the time by the thermostat it's on.

Ick.

Mining on laptop BAD.

Good way to kill it quickly from overheating it.

I'm amazed that has enough video RAM available to mine ETH at all (My REAL GTX 960 cards are 2GB, which isn't enough to mine ETH any more).

I was stationed on USS Ranger out of North Island for a while - there were parts of the WestPac I made I definitely would have liked to repeat (like the stopover in Thailand) and other parts I would like to have missed.

 Chula Vista- varied when I was living there, some parts were decent some were starting to get iffy.
 Nowhere near as bad as Barrio Logan though.

 Climate in Central Washington isn't a ton different, except we actually have winters here (usually mild though).


 I wish I had the money for 50 x 1080 ti cards AND the rigs to put them in. Might get into that ballpark in another year, if profitability stays good, but I suspect it's going to take a good bit longer.

 I like my EVGA SC - probably the best 1080 ti that does NOT need 2 x 8-pin power.

 Gigabyte Aorus at 175 watts was also about 685 sol/s but only runs about 60 C for the one in my gaming machine (80F ambient in a well-cooled case).
 The other one I have in a mining rig (82F ambient) is running 220 watts 745 sols at 50C (the fans at 80% are definitely overkill and I should turn them down a lot).
 (edit)  That's at +100 core / +100 memory, higher memory settings gave REDUCED hashrate (I think the Aorus "overclocks" the memory by default), higher core was long-term crashy.

 Viable if you run them at about 230 watts or less (180 ballpark for the power supply that is ALSO running the MB/RAM/HD stuff) - and your power limit setting is RELIABLE (which is not always the case on Windows, Afterburner among other options will sometimes LOSE settings on one or more cards after a driver reset).
 I wouldn't push them much if any harder than 700 watts at the wall even on a GOOD power supply, for long term reliability reasons - even with the TDP turned down, you still get SPIKES on power draw quite a bit above the set point.

 On an overall "system cost" basis if you are aiming for max efficiency, the 1070 ti beats the 1080 ti on a hash/$ basis AND on a hash/watt basis - but the 1080 ti system will have a bit more ability to expand hashrate *IF YOU HAVE ENOUGH POWER SUPPLY CAPACITY* and are willing to live with lower efficiency - and the figures come out pretty close, you're not going to see a 1070 ti based rig making money and a 1080 ti rig losing money unless you're pushing the 1080 ti cards a LOT harder.

 At this point, on both the 1070 ti and the 1080 ti my "go to" card is the EVGA SC version.
 I like my Gigabyte Aorus 1080 ti cards, but the dual-8 pin power connector makes them a headache to get multiple cards hooked up compared to the 8+6 on the EVGA SC, and since I'm aiming for good efficiency the cooling on both cards is plenty and they perform equally 'till you push them up past about 220 watts (the better cooling on the Aorus lets it boost higher at that point for a bit higher hashrate and efficiency).
 The SC also tends to cost less than the Aorus, but on a $750 ballpark card the $10-$20 usual difference isn't a big deal in and of itself.

 5 x 1070 ti also fits into the power budget of my rigs, where I end up having to run only 3 x 1080 ti and set them to a somewhat less than best efficiency point (and STILL end up with less total hashrate).
 Down side is the 5 card rig WILL take up a fair bit more space - but even there it's not an issue in my specific case as my current "shelf/rack" design fits a 5 card rig very nicely on a single shelf with good cooling, but would be overkill for 3 and WON'T fit a pair of 3-card rigs on one shelf without major cooling issues.

Never looked at or heard of sharkmining before - and as a computer tech with experience going back to CP/M and S100 days (that's BEFORE IBM made their original Personal Computer, for the young folk among us), I'm not inclined to look at prebuilt machines at all.
I do rare exceptions for ones that are being sold for less than the cost of usable parts, like my "Big 3" rig I got from a NewEgg closeout that started out as a pre-built for Scrypt machine with 3 x R9 280x - and the value of the video cards plus power supply ALONE was more than the cost of the machine, leaving the Honking Huge Massive Cooling case and MB/CPU/RAM/HD as "bonus".
It has 3 R9 290 cards in it now (the PS was severe overkill for 3 x R9 280x) and I swapped in a dual-core for the original Sempron 145 at one point when the MB the dual-core was originally on died, but it's otherwise the same as I originally bought it (the R9 280x cards ended up in a different machine).

 EWBF does in fact run on AM2 motherboards - I've even run it on Socket 754 based Sempron systems which are a few years older than AM2 at times in some of my testing.

 I'd be more inclined to look at Windows 10 or drivers as the issue, especially since Win 10 was not designed to run on anything quite that old and needs MASSIVE resources compared to previous Windows versions.

 Unless it changed recently, it DOES in fact check driver version - and mines in a "slower" mode if you don't have the blockchain drivers.

 I have no experience with the Duke, but IIRC it needs 2 x 8-pin connectors which makes it harder to run a lot of them in one rig.

 EVGA FTX is the same way as I recall.

 SC2 is 8-pin plus 6-pin, has good cooling (though not quite up to my Aorus cards which are ALSO 2 x 8-pin), and runs quite well - also seems to be the highest-performance card that does NOT need 2 x 8-pin connectors at this point, though that's splitting very fine hairs.

 Very nice - best I've seen anywhere in the US in the last decade was $1000 for a 3000 square foot "mostly warehouse with a small office space" setup I had in the Indianapolis area for a while, unless the place was HUGE (10,000+ feet).

 $500 for 1000 feet with water/sewer/garbage included where I'm at now though is good for this area, expecially with the "cooling vents/louvers" built into the building front-and-back of the space, the "gable-type exhaust fan" that was already installed in the space (but in a bad spot, it'll work better as another intake), and I'm on the end of the building with 2 A/C units mounted in the "side" wall (one I converted to additional "vent" area with prior landlord approval but in a way I can easily put it back the way it was when I move out).

 I suspect part of the reason it's as low as it is would be the ACTIVE railroad track (single-track BNSF main-line) the property backs up onto - but I grew up almost as close to a dual-track Penn Central line so the train noise is no big deal to me.

 A standard 192.168.x.x privatenet offers ballpark 64,000 usable addresses - I don't think even BITMAIN needs much more than that, though I'd guess they probably use a 10.x.x.x privatenet "just in case".
 For example, I run my A2 miners on the 192.168.2.x subnet, most of the rest of my machines on the 192.168.100.x subnet, I USED to run a bunch of machines on 192.168.10.x via 10Base2 coax, I USED to run my S5 farm (and the SP20) on the 192.168.5.x subnet, and I have a DHCP address space set up for "short term usage on new machines I've not had the chance to fully configure yet" on 192.168.254.x
 My mining farm area though is starting to subdivide - each 6-rig "rack/shelf" unit will have it's own 192.168.100.xx1 through .xx6 address set, moving probably to 192.168.0.xx0 range if I get big enough eventually - makes it easy when I look at remote monitoring to say "ok, that IP is down, it is located THERE".
 I already had that implimented on the previous shelving units, each "row" with it's own .xx# and the # being the rig IN the row from top to bottom.
 I'm SURE this is not a new idea, though I've not seen it in actual usage anywhere that I can remember.

 It's NOT a good idea to push 300+ watts through a "dual PCI-E" type cable in a lot of cases - check the gauge of the wiring first, some of them are only intended to handle the 225 "8-pin + 6-pin" that are attached to them with VERY LITTLE left over if at all.
 EVGA and Seasonic cables should be OK - EVGA uses fairly heavy duty wiring on it's VGA cables for the G2/P2/T2 series (and I think they carried the cables over to the G3 series) and Seasonic tends to "double up" on the wires on their dual-connector cables.

It seems likely you are running in the the Ubuntu BUG in one version where xorg.conf gets DELETED at bootup - the real workaround is to add " NOGPUMANAGER" to your grub startup line in /etc/default/grub then update grub.

You keep forgetting to post the link on where to download the pretty woman.

 9-)