ETH mining on the GTX 1070 ti is a waste - they do a LOT better on equihash algorithm coins among other choices.

 The reason the TDP isn't having much effect is that the ethhash algorithm is VERY memory-limited, so dropping the core clock on MOST cards doesn't affect the hashrate for a long time as the memory can't FEED the core enough data to keep up with it.
 Even the R9 290/390 with their 512 bit memory bus can run the core at noticeably "lower" settings than on most coins and still hit max hashrate on ethhash.

 No current motherboard I am aware of has the 1-4-7 slot spacing to do what you want to do and have fairly good cooling to all cards without going to rather expen$ive "5+ slot" boards - and very few of THOSE put a 16-bit slot in that #1 position.

 When you start looking at paying $300 ballpark AND UP for the motherboard, you quickly start looking at other options.

 Buying the chips themselves is not a problem - they flat out ARE NOT AVAILABLE PERIOD unless you disassemble an existing S9 to get them (or R4 or T9).

 As has been mentioned more than once already, BITMAIN DOES NOT SELL THEIR CHIPS.
 PERIOD.
 NOT AN OPTION.

 If you see BM1387 chips for sale, they are either "pulls" or they are FAKE.

 Since Bitmain doesn't sell the BM1387, they don't see any point in making specs and pinouts and such information available.


 I am often a bit shocked that Bitfury still sells THEIR chips - but I suspect it's inertia as they've been working with "major mining farms" and just selling chips to THEM for years.

I'm pretty sure Apple does a lot of it's manufacturing in China.

Most MODERN power supplies automatically accept input from about 100 VAC to about 240 VAC with no "switching" needed.
Only older "transformer-input" type supplies needed the switch.

 US outlets for 234VAC circuits are NORMALLY NEMA 6 series outlets (this is NOT the same as the common NEMA 5 outlet you see in most houses).
 Cords for NEMA 6 to connect to standard power supply connections tend to be less common, and more expensive, than the very common NEMA 5 cords - there are quite a few folks that sell them though, check eBay, Amazon, Newegg for them.

 The cord you have the link to IS the type you are looking for - but way expensive, and you don't really need a "heavy duty" version as you're pulling well under 10 AMPs to fully power a APW3++ at 234 volts.
 The cord shown appears to be a NEMA 6-20 (20 amp) cord, you're more likely to want to use NEMA 6-15 cords and outlest (a NEMA 6-15 plug WILL plug into a NEMA 6-20 outlet, but not the other way around - like on NEMA 5 plugs/outlets the 20 amp version has one "T" type socket connector).
 With some digging you should be able to find applicable cords in the under $10 range.

The older fglrx drivers can't be used on Ubuntu past 14.04 due to changes in the kernel that the drivers don't work with - and even 14.04 was a nightmare after Canonical "updated the kernel" to a version that did NOT support them (this can be reverted via apt, but it's a PITA and Canonical caught a LOT of flack for changing the kernal version and BREAKING a lot of stuff without changing the RELEASE version).

Neoscrypt is widely reported to be "memory heavy" like ethhash and cryptonight.

Cryptonight currently seems to be mostly mined with Cast or Stak-XMR - ccminer and Claymore have both fallen noticeably behind.
Equihash - add BMiner
Ethhash - I don't think anyone has used ethminer for a LONG time, mostly Claymore and some diehard Genoil fans, though there is a new "phoenix" miner out for Windows that I've not been able to check out yet.
Qubit - Baikal ASIC for over a year now.
I think the Baikal units also support Groetsl or Skein, possibly both, on their most recent design.
X11 - MULTIPLE folks making ASIC for over a year now, including Baikal, Pinidea, IBeLink, Bitmain, and Innosilicon
Scrypt - MULTIPLE folks making ASIC for 4+ years now, currently Innosilicon, Bitmain, and BW.com

Equihash - depends on the card to at least SOME degree, and sometimes different models of the same GPU work differently depending on where the FACTORY CLOCKS are set.
 Usually wants some OC on both core and memory, but I've got several 1070 cards that don't see ANY benefit to memory OC on that algorithm.



 Just got around to my daily check on Newegg - and OUCH, RX570/580 pricing just went through the roof and is now obviously pushing NVidia pricing.
 ONE model under $400 (yesterday still had a couple under $300) and 1070 pricing pushed WELL into 1070 ti pricing on the very FEW available cards for EITHER side.

Stability of a non-riser trio rig can be a bit better - fewer parts to go bad - but COOLING of the middle card in such a rig is a major problem.
You can work around that somewhat by using a "shorty" card as the third card, like a Zotac mini or the Gigabyte ITX models, but then you get to deal with the junk fans on the shorty cards (EVGA makes short cards with good fans but I don't think they go above a 1060 on them).

 I built quite a few such rigs during my Folding days, as Folding@Home needs a lot of CPU support and a lot of PCI-E bandwidth to get good PPD (hashrate) out of the cards so riser rigs were NOT a viable option.
 In theory, RIBBON-type x16 risers would have worked, but those tend to be a NIGHTMARE to work with and also tend to be very short.

It's about the wiring, not the PCI-E spec itself.
I don't think they anticipated folks using wiring runs long enough for IIR losses to cause issues, or enough losses to cause noticeable voltage drop to the connector.

 I'm not quite THAT close to power limits - be a while before I have to worry about that.

 I DO tentatively plan to replace older cards with upgraded ones at some point though - especially the cards that are "lower efficiency" NOW - but I've got 5800 watts or so of A2 ASIC that's getting kinda marginal again to "replace with something" before I worry about replacing older GPUs, along with a few GTX 9xx series cards.

 As a general rule, if you look at your main breaker box, the BIG breaker at the top will be the "input" breaker and will tell you your total capacity for the house as a whole.

 Most common house outlets are 15 amp - but check which ones are on which circuits, most houses have MULTIPLE outlets per circuit on many of the circuits, and usually room lighting on the same circuit as one or more of the "in that room" outlets.
 I've seen a LOT of houses where the wiring had outlets in DIFFERENT rooms on the same circuit - the last one I lived in had one circuit that put outlets in 3 different rooms AND room lighting for one of those rooms, but that was an old house with some weird wiring and major issues anyway.

 Breakers are marked with their capacities, usually on the handle.

 DO NOT ASSUME 120 VOLTS.
 The actual "nominal" voltage in the US is 117 VAC, and it's common to see anywhere from 110 to 119 VAC - I've never actually SEEN a circuit deliver 120 VAC at the outlet, though my current place hits 119.3 or so intermittantly and seems to average right about 119.0

 The issue isn't on 20 amp circuits, the issue is the COMMON 15 amp 117 VAC circuits most US homes and small businesses have, AND that the S9 isn't actually specified for "1300 watts" but for "1323 watts with up to a +10% deviation" on the recent 13.5 TH batches so in some cases it's SPECIFIED to use 1455 watts.

 20 amp circuits on 117 VAC are NOT "typical", though they aren't rare especially in kitchen areas.
 Even 20 amp CIRCUITS will often use 15 amp OUTLETS - 20 amp outlets ARE uncommon on 117 VAC circuits except some that are intended for use with a high-capacity window air conditioner.

 15 amp circuit at a "low" 110 VAC (which is WELL within US power distribution specs) allows 12 Amps continuous operation = 1320 watts, which is marginal for a S9 that is at the BOTTOM END of the specified power usage range.

 Running a S9 on a 20 amp 117 VAC circuit isn't an issue - 16 amps at 110 would give 1760 watts which is quite a bit more than even the highest-hashrate batch is specified to pull INCLUDING the "+10% deviation" part of the spec.

As I recall, AMDGPU-PRO drivers don't support anything older than the Polaris (RX-series) cards.

One of my MANY ISSUES with AMD the last couple years has been their FUBAR older-card support on their more recent drivers - especially for stuff like the A10 iGPUs that even under WINDOWS have very iffy support since 15.12 despite still being current production at the time.

 LINUX has been MUCH worse - one of the reasons I've moved AWAY from Team Red since 2015 has been the total GARBAGE that AMDGPU-PRO has turned out to be.

They need to be less expensive than the 1070 to be worth buying, as it appears they are nothing more than a 1070 GPU coupled with INFERIOR (for mining) GDDR5X ram and only 4GB of it (the lower memory amount is not an issue for mining).

 At least one maker CLAIMED 35 Mhash/sec on ETH but the use of GDDR5X memory tends to make a card SLOWER on ETH mining, so I call BULL on that claim - or they had a typo and meant to say "25 Mhash/s" (which WOULD be believeable as that's close to what the 1080 can manage with the SAME type of RAM).

 I would expect their performance on non-memory limited algorithms to be the same as a 1070 at the same core clock, give or take "silicon lottery" factors.

Try running it on the Nicehash Legacy miner.

 They're 2.x series mining software seems to have a LOT of issues and flat out doesn't work for a lot of folks.

 Samsung doesn't have "higher initial performance", it's all about the common ability of it to overclock more than Micron or Hynix with reliability that makes it the better option for mining.

 I pull enough coin to pay the bills as fiat, the rest gets reinvested into more hardware to grow the farm.

 This might change at some point in 2018 as I'm starting to see the limits of my infrastructure - worst case I shut down the A2 farm to free up some additional power for the GPU side, though I've also been thinking about replacing the A2s with 841s or something.


 Anyone else notice that NVidia "good mining" cards seem to be in much worse shortage right now than AMD "good mining" cards?

 Even the 1080 ti is getting noticeably short - which did NOT happen during the big Spring mining surge.


 I have to wonder if NVidia is ramping down Pascal production in favor of ramping up Volta production so they have a "good supply of cards" on hand when Volta gets announced and goes on sale (possibly Feb, probably March timeframe, per a comment an EVGA employee dropped somewhere in their forums).

 Get a bigger house, or a "mining cave".

 9-0

 How many card models support CUDA 9.0 to date?

 Also have to wonder if it would offer any significant performance increase, IF at all - sometimes a "CUDA" compute upgrade is about adding FEATURES not better efficiency.

Doesn't your pool support ZEC itself?