For Bitcoin itself, the most efficient miners are the S9 and R4 (equal more or less depending on specific S9 batch and individual unit varience), then the T9, then the Avalon 721/741, not sure after that.

 There are BitFury models entering the market but they're new enough the actual performance is yet to be determined reliably.


 With that said, mining Altcoins to convert to Bitcoin can be a lot more efficient in many or most cases right now - especially if you use the Baikal ASIC miner or possibly the Bitman L3+

Under 1% isn't enough of an issue to worry about.

 You are ALWAYS going to have occasional shares rejected 'cause someone else found it just before you did and sent it in first, due to Internet Latency.

 Higher latency = more "stale" shares lost like this.



 As a general rule of thumb - you should anticipate your "stales" rate being in the ballpark of "your latency / time between blocks".

 Also keep in mind that the latency depends on the pool side as well - a pool suffering a DDOS attack will see it's latency go up as a result, sometimes by a LOT.

Interesting stat here.

https://blockchain.info/stats

 Percent of income from transaction fees = 19%

 Looks like the "when the block reward runs out" issue some folks worry about is ALREADY getting eliminated pretty quickly.

And what TEMP is the room with the A/C at?

 The curve of how much water a cubic meter of air can hold at a given temperature is NOT linear vs temp - it is a curve that goes up quite a bit faster as the temperature increases, with a MAJOR spike upwards at boiling, and a major drop at freezing.


 If your A/C room was at the same temp without the A/C dehumidifying the air, it would be probably in the 80-100% RH range.

 Electric cost on a hard drive is trivial. I can't think of one I've seen that ate 10 watts for a while, and can't remember the last one I saw that ate 15.


 I'm sure the SMR drives plot slowly - but you only have to plot ONCE, and you don't have to plot the entire drive as a single plot file.
 I'll speak with more authority about that in a couple weeks, I've got an Archive on order but it doesn't show up 'till next week - probably going to put it in a new system build, but I might put it in one of my older existing systems instead.


 I do agree with "go for the lowest cost per TB" though - and Europe prices seem to be real spotty compared to US pricing, probably due to less competiton and higher pricing in part due to the VAT that many (most or all?) EU countries seem to love.

 Asteroids@home - determine the orbit of asteroids in the hope of figuring out the next "dinosaur killer" or even another Tunguska even that ISN'T in a very very low pop area soon enough that we can DO something about it.
 (edit - oops, misremembered this one, I was thinking of a different project this one is about figuring out what an asteroid's configuration is and perhaps a bit about it's density)

https://en.wikipedia.org/wiki/Tunguska_event

 On the other hand, in the last couple of decades astronomers have found a LOT of "potentially habitable" worlds, giving a very strong point to the Fermi paradox.

 https://en.wikipedia.org/wiki/Fermi_paradox

Takes a VERY VERY high end GPU to even think about plotting 1 TB in an hour.

 On the other hand, you can keep GPU mining while you plot if you use a CPU-based plotter and leave one of the CPU cores available to run the GPU mining with.

 3 cpu cores of one of my A10-5700 plotted 1 TB in under a day WHILE that rig was mining ZEC on a pair of HD 7870, no issues at all.

Best bang for the buck has to take in the total system cost though, not just the cost of the cards themselves.

 That's where the 1060 takes a noticeable hit - it's efficient on a "per card" basis, but the total system cost to hit a given hashrate needs quite a few more systems so the "per card" basis ADDING IN THE ADDITIONAL SYSTEM COSTS gets a lot closer.

 It might still be "the most efficient" overall but it's not going to be by a huge margin if so, unlike if you only count the cost of the card itself where it is pretty dominant on ZEC vs the 1070/1080/1080ti.

 You also left out the 1060 6gb - which has been reported over 300 sol/s with EBWF but doesn't cost a lot more than the 1060 3GB models.




 to Bakes007 - the rates vary some on ZEC mining because the cards themselves vary (FE cards might be an exception among themselves).
 There is no such thing as "any 1070 can do any specific speed at any specific wattage used" because of variations in how the third parties build the cards, and the parts they use, ESPECIALLY MEMORY.
 For a specific example, my Gigabyte 1070 full-length cards (180 watt TDP) have "stock" clocks noticeably higher than my EVGA 1070s (all long card, 155 watt TDP) or my Gigabyte "ITX" short 1070s (155 watt TDP), and seem to cool better, so they can be pushed up to 430 sol/s fairly easily where the EVGA cards and the ITX Gigabytes have to push hard to get much over 400 sol/s - this is in systems where the rest of the system is IDENTICAL (in fairness, the ITX cards always get inflicted with a PCI-E 2.0 "16x" slot that only has "4x" lanes wired to it, but for most cryptocoin mining that's NOT an issue or "riser rigs" wouldn't be so common or work worth beans).

 Memory brand also has an effect, as some brands will let you clock the memory a LOT higher than others (more of an issue mining ETH but it's noticeable on ZEC).

 My only 1080 is also a long-card Gigabyte, and can push 500 sol/s pretty easily. Tradeoff is the cost - $489 for that one ($499 for the second one I just bought) vs $379 or $389 for any of my 1070 cards.
 When you factor in the cost of the total number of SYSTEMS and the total SYSTEM cost, the 1070 and 1080 get a lot closer on hash/$.


 I too have noticed the efficiency of the different cards changes from one coin to the next - the 1080 I have can do quite a bit over 600 MHhash/s on Skein, the 1070 Gigabytes might manage 500 or so and have to push harder to get there, which on a raw card basis gives the 1080 slightly better hash/$ and even more so at the system level.
 On ETH, though, the 1070 Gigabytes and 1080 Gigabyte have almost identical performance near 30 MH/s, indicating that the extra cores on the 1080 are barely enough to overcome the higher latency of the GDDRX 5 memory on the 1080 (even though the memory speed is HIGHER, the latency is worse which is a MAJOR issue on ETH) - and the EVGA cards only lose 0.5 MH/s at most to the long Gigabytes despite the almost 10% lower core clock on the EVGAs.

 The reason AMD is dominating on ETH is that you can match the performance of a GTX 1070 almost identically with a minor BIOS mod on the 480 (there are no tools for BIOS modding a NVidia 1070, other than trying to flash the card with a BIOS from a different model 1070) on hashrate and VERY close on power, while paying ('till the last week or so) a bit over HALF THE PRICE for the card.
 Even bone stock, my pair of Sapphire RX 470 will manage 22 Mh/s on ETH - and only cost me about $170 each new from Newegg (I got them to match the deal on their own EBay site under their price match guarentee but they would have still only been $183 or so otherwise WITH shipping) - which makes them a TON more efficient on a Hash/$ basis than any GTX 1070 has ever been.

 Bitcoin price does not care about the difficulty.
 Bitcoin mining profitability is the ONLY thing that does.


 TSMC is talking mid-to-late 2018 for 10nm at the soonest (and that would be very small quantity and very POOR yield).

 They might have a "NVidia Custom" 12nm process before that, per a few rumours about the potential NVidia 2xxx series "refresh", but no hard information on that possibility and anyone but NVidia getting access to that tech would come quite a bit later (and the process is rumoured to only be a SMALL improvement over existing TSMC 16nm process options).

  Again, 2019 at the soonest for anyone other than the "big players" to get access to a new node with a significant performance jump.


 At current price and difficulty, profitability is quite a bit HIGHER on a S9 then it was when they were first released - despite the large diff increase since then.
 There is no need for "new miners" anytime soon to maintain good profitability.

 

BURST profitability is a lot higher NOW than it was even 2 months ago - despite the network "TB estimate" having more or less doubled in that timeframe.

 3 months ago, I was figuring out how long ROI would be on a Seagate Archive 8TB - and came up with almost 2 years.
 I did that calculation early this week - and came up with a low single-digit MONTH figure - which caused me to actually pull the trigger on buying one to stick into the "new rig" I'm going to be building up when the parts arrive next week.


 For someone with severe power limits, BURST is probably the highest income/watt coin in existance and has been for a while.


 I dread the concept of BackBlaze deciding to use BURST to test their new-built "pods" and "vaults". 8-O

 Having used a lot of Seasonic PS over the years (including some G-series), I have to say that even Seasonic has an occasional "infant mortality" issue.

 Plus side - if it's infant mortality, you generally have no issue getting a quick exchange through the dealer you bought it through.


 A G-450 should be plenty for a single 1070 based rig - I've a couple of 2-card rigs running on the G-550 (though one of those is a pair of HD 7870, but the power draw of those cards is still 140+ watts each running MooWrapper and highly overclocked) and my "go-to" supply for my 3-card 1070 rigs is the X-850.


 Ignore the Prime series though - Seasonic decided that they were going to prioritise "quiet" over "long term reliable" in those and went with a bloody SLEEVE BEARING fan instead of their long-standing use of double ball bearing fans.

 If you don't like the dev fee, there are other options.

 The only thing to worry about on the 1080ti are the rumours that NVidia might drop prices some on the 2xxx "refresh" when it shows up - but that appears to still be months off, you should make plenty to make up the difference by then even if you don't manage to pay off the 1080ti in that timeframe.


 It looks like I timed my last "rig parts" purchase near perfectly, as BTC started sliding some a few hours later.
 8-)

 (edit) and WOW has it slid a lot since I made that purchase.
 So nice to be psychic for once!

 8-)
 (/edit)


 Even if all cryptocoin drops 50% over the next month (edit - that's starting to look POSSIBLE, I was just doom and glooming when I originally posted that comment), most miners should STILL be way ahead on profitability over the last half of last year and the first couple months of this year - and those that didn't use the recent "surge" to finance expansion are just losing out (unless they're severely power-limited, in which case they should have been retiring older less-efficient ASIC and GPUs and liquidating them in favor of newer more-efficient ones).


 DGB profitability has been bouncing all over for the entire week-and-change I've been tracking it - way too small a total network hash on ALL of it's options, even a single mid-range miner like I am can have a noticeable effect on it.
 

 The AORUS cards are beasts - very high clocks, very high-end cooling for air-cool cards - if you can handle them being 2.5 cards wide and the needed spacing they look like the best option on the high-end (water-cooled stuff gets too expen$ive to be cost effective, with the POSSIBLE exception of those folks in hot+wet climates that can't use evap cooling to get the hot down to a tolerable level).


 IRS compliance being "binding" has been challenged in court a few times - with ZERO successes.

 At this point the applicable precedents are all "you WILL pay taxes or you WILL pay the penalties or you are a criminal".
 While precedent is not BINDING on a judge, it is a very rare case that a judge will rule AGAINST an existing precedent where there are ZERO contradictory precedent to work with and quite a few in favor of a particular interpretation.

 BURST and Seagate Archive drives (or any other SMR drive) are a PERFECT fit for each other.

 Archive drives are designed to be written to rarely, read from a lot more - which is exactly how BURST works as you plot ONCE then read lots of times.

 They are also THE lowest $/TB of any drive type on the market.

 Amazing how they fit your "if I where (sic) you" line perfectly yet you deride them.

 That's why you bump up the fan profile - ALL recent video cards prioritise "quiet" over "cool", with the early EVGA 1070 models being a POINTED example in going way too far that direction to the point they were having some of them fail due to overheating BECAUSE the fan profile had the fans running so slow they couldn't respond to load spikes fast enough.

 1080ti tend to have better fan profiles to start with because the makers *KNOW* they are going to generate a lot more heat so they bumped the fan speeds up a lot higher to start with than on a 1070 - and if you pay attention you will note that most makers have the SAME cooling solution available on their 1070 models as on their 1080 and 1080ti solutions, except for some of the LOW END 1070 models with cooling solutions that can't handle a 1080Ti at all.


 IMO any miner that doesn't have the fans on his GPUs running at 90% or more by the time the card gets to 80C (and preferably 70-75) is DEMANDING to have their card overheat and die.



 As far as "most 1070 can do over 31Mh/s" - I call BULLSHIT on that one, NONE of mine have ever managed more than 30 no matter HOW HARD I pushed clocks, even at 180 watts much less 95.
 ETH depends WAY too much on low memory latency, and that VARIES WIDELY depending on the brand and sometimes the specific chips of the memory in a card.


 Mining ETH on a 1070 isn't a lost cause.
 It IS still very profitable, much more so since the recent ETH price runup, just LESS so than other options.

 
 NVidia-SMI is nice, but I've seen it get confused and "off" often enough that I don't trust it to be entirely correct.

 Given the totally unusable so-called "user interface" on 8, you can't PAY me to try to use that turkey piece of junk in a mining machine ever again.
 One attempt, and never got past the damned default so-called UI in 6 hours of TRYING was more than enough.
 It took me less than an hour on that SAME machine (that came preinstalled with 8 and SUPPOSEDLY already had miner software installed) to blow 8 off it, install LINUX, install dependencies, compile and install and configure mining software, and get the machine hashing.

 It says a lot about 8 that it looks like Microsoft is going to support it for a SHORTER total period of time than they offered support for ME or Vista (both of which got outlived by YEARS by previous Windows versions).

 When even Microsoft concedes an OS version is junk......you KNOW it's got to be horrible!

 No "so to speak", A/C units ARE dehumidifiers and vice versa - if you look at almost any recent window A/C unit it will have specific "dehumidity" settings on it.

 That is the reason A/C coils will literally ICE UP sometimes if you don't allow enough airflow to the condenser coils.


 Higher ground will *sometimes* have lower humidity - depends on the details for your specific area - but it's usually not by a lot even when it is the case.


 Humidity IN your mining room will be lower because of all the heat generation raising the temperature - which lowers relative humidity because the capacity of the air to hold moisture increases as the temp rises.

If you use your CPU to mine BURST with while mining on your GPUs, you won't have any issues other than that CPU miners are slower so you might occasionally miss a deadline on a "fast mined" block.

The GPU-assisted BURST miners can work as well, but they get a LOT slower when you're also mining something else on that GPU.

 Took me less than 2 days (more like 30 hours IIRC) to plot most of a 3GB drive with an AMD A10-5700 - JUST using 3 cores of the CPU.

 GPU plotting can be quite a bit faster if you have a mid-to-high end recent GPU though (I tried it with one of my HD7750s, that was SLOWER than using the A10 as I recall though not by a lot).


 One trick that helps some - plot smaller files at a time, then you can mine while you're plotting on the completed plot files.

I'd bet late 2019 at the EARLIEST - even AMD isn't talking 2018 for 7nm and they are one of the 2 PRIMARY customers for GF.

 More likely 2020.


 Can't build miners on a node that is still in the "experimental testing" stage.