^But right at this minute we have to consider capital cost, so it's not just hash/watt but also hash/euro or hash/dollar invested (there's a difference).  

Remember the football book, "One knee equals two feet"?  One GTX 750 ti equals two RX 480's, with the main cost driver being how many of those suckers you can cram onto a node.

Linux infrastructure for a four GPU node is about $100 (I use HP server PSU's to keep the cost down).  Windows + 6-slot motherboard alone will run you like $180 at a minimum, not counting the extra RAM and HD.  

If I had to get into GPU mining right now today, I think the older cards are winners.  Never tried a 750, but the numbers work right on a Linux system.  Believe it or not the 6850 deserves a look for the same basic reasons.  

Only thing is, the hashrate data on 750ti is kinda old.  And I don't think they have enough RAM to mine ETH (do they?) and technically if you're gonna do GPU mining, these days it's more profitable to mine ETH (and on some days, ZEC) and flip it to XMR or whatever you want.  

290x can churn and burn any of those things.  Clear winner for some European miners due to low cost of electricity and ready availability of 220/240V (because you can squeeze a little more juice out of some PSU's, and run more machines without burning the house down).

The math isn't that complicated, but you can't just eyeball it, gotta run the numbers. (That's why God invented OpenOffice, and no Bill Gates is not God.)

The 290x is a monster, great card, uses a lot of power though.  I get like 740h/s out of it, compared with 650 for my 480's.  Love that card, but it's best if your energy costs are subsidized.

Advantage of linux is
- low hardware overhead
- great for networking
- easy to run headless
- reliable
- free

Advantage of windows
- more tools in windows to optimize GPU performance
- more compatible with some modern motherboards
- less learning curve initially (I guess.  I wasn't eager to maintain a windows network)

Consider old, refurb business or workstation motherboards.  HP makes good ones, just make sure they have standard power connectors.  Old is good for linux, you really don't want to have to deal with UEFI if you don't have to.  

That said, I've had good luck with MSI UEFI MB's.  They tend to come from the factory with the right settings.  Still helpful to add "iommu=soft" to your kernel parameters (google it up).  If you don't, sometimes half your USB ports go dead, which doesn't bode well for your USB drives.  You don't run into those issues with older pre-UEFI MB's.  

Long as linux plays well with your MB, USB drives work fine.  Check out Clonezilla, which is a nifty linux timesaver.

I like AM3 sockets with inexpensive, low power sempron or athlon chips.  2g of RAM is enough.  

Tricky part about mining with the 290x is choosing the right power supply.  For a 4 GPU rig you'll probably need 1200w to get enough overhead, and enough connectors.  The 290x is such a beast, there's an argument for running 3 to a motherboard.  Get by with a less expensive PSU.  Just depends on what your various components cost.  Gotta do your own math but the bottom line is, don't skimp on your power supply, or on your risers.  Get the best you can buy, don't burn the house down.

290x is a great card for XMR but to maximize profit, consider mining to an exchange, mine whatever is most profitable at the time (ZEC is looking good these days) and then flip to whatever you want to hold, whether it's XMR, BTC whatever.  

Xubuntu 14.04 (trust me; I suffered so you don't have to).  Set up fixed IP address and DNS server ( 8.8.8.8 ) make sure internet is working.  Turn off updates once you set location and language.  Radeon drivers are in the repository.  Install your graphics cards one at a time.  Compile your miner. Install SSH server, set up public/private key authentication, turn off password authentication.  Get into grub and set it up so you boot to the command line.  Shazam!  I love this ****

Linux isn't too too hard.  Once you start setting up your network and SSH you'll see the beauty in it.  Xubuntu 14.04 is a straight up vanilla, lean linux install that works like a charm right out of the box.  It practically installs itself.  Figure out how to use the Debian/Ubuntu repositories.  There are some excellent articles on setting up SSH over on the Ubuntu forums.  After that you'll find the best documentation over on the Arch website, believe it or not.

OpenCl goes in /usr/lib.  Just sayin.  Someday you'll thank me for that.  

There's probably an opportunity in the US for the next few years due to some peculiarities in the law

When I'm out of town I set my machines to one pool and then monitor that pool over the internet.  Controlling is another matter.  Linux is like hitting the easy button (although that is literally the only thing that's easy in linux).  You remote in to linux using SSH, which is a simple terminal program that has some really cool (and convenient) security features built in.

They do make SSH servers for windows (eg Putty) and clients for smartphones and whatnot.  If you're cool with the command line, that's one option.

Earlier this week, Jim Rickards commented in passing that Bitcoin, gold, and USD are all forms of currency.  None has a coupon; none is an investment.  A pyramid scheme refers to a fraudulent investment scheme.  So it doesn't apply to bitcoin.  

Beyond that, the whole notion of currency as a store of value is nonsense.  If currency had intrinsic value, it would be barter, not cash.  Krugman's position -- that the dollar is a store of value because the US Government has lots of guns -- speaks to the government's ability to enforce USD as a medium of exchange world-wide, but says nothing about store of value.  Back when USD could be exchanged for gold,
 you might have had an arbitrage opportunity, depending on where the face value deviates from the market value, but trading dollars for gold,
 even then, was a matter of trading one currency for another.  (Now, where's my Nobel prize, dadgumit.)

Marks, an investment adviser, doesn't feel bitcoin is a good investment.  He's right, by definition.  It is possible though (I'm told) to make a living speculating in currencies (Soros did well for himself eh).  Options:*

1.  Try to identify, and exploit inefficiencies in the marketplace.  In other words, buy the dips.  Good luck with that; if that's your strategy, seems to me like now is the time to sell (for the reasons Marks details.)
2.  Run an exchange (probably the best option).
3.  Bank on the real purchasing power of BTC as opposed to say USD at some point in the future.  In other words, how much stuff will I be able to buy with 1BTC a year from now (if anything), compared with what I'll be able to buy with $2600.  

The latter strategy, which is probably what most of us are banking on, isn't precisely an inflation hedge.  Inflation only looks at one side of the equation, namely the supply of money.  Money, like everything else, is governed by the laws of supply and demand.  What keeps USD afloat is the almost insatiable demand that comes with being the world's reserve currency.  

If you're hedging against the possibility that the USD falls from that lofty perch, you might should be stocking up on canned peaches and toilet paper, rather than BTC. Otherwise, the real question is, what's the emerging demand for BTC.  And that's kind of hard to tell.  

For example, what explains the recent surge in BTC/USD?  Is it just a "better fool" contest?  Or is demand surging?  There have been two events in recent months that could drive demand.
- India's war on cash
- Hot money coming out of China, evidently spurred by worries about devaluation, seeking to dodge capital controls (check the article on real estate values in Seattle, earlier this week on ZH)

OTOH, a couple of recent busts on the dark side show just how transparent the blockchain really is.  That should decrease demand.  I don't think people care about "store of value" as much as liquidity.  But, I think most of our customers care a great deal about confidentiality.

I don't think now is a great time to trade USD for BTC.  If I held Yuan I might feel differently.  But in the long run?  I don't think we are anywhere close to seeing what the demand for this kind of liquidity will prove to be over the coming years.  

It's OK to disagree, of course.  What do I know.  But a pyramid scheme?  Cmon.  It's not so much that Marks is a dinosaur, or an economic illiterate.  He's just a troll.  


---------------------
*Mining offers some arbitrage opportunities, but that's off topic

I don't think so.  Run the numbers and you'll see.  I think the main advantage of GPU is flexibility, you're pretty much stuck with XMR if you want to do CPU mining. So you're coming in at a high difficulty level with an important inflection point coming up, in terms of issue rate.  But yeah, if you're saying we are at a unique time, I agree.  Convergence of high XMR value, lots of servers coming off-lease, GPU shortage, and a brief time left to mine up some XMR profitably.  

Don't know whether to thank OP, or remind him that the number one rule of fight club is, you don't talk about fight club.

If you really want to get a handle on the numbers -- and see why this is a great idea -- you need calculate the net present value of the project.  NPV is money in minus money out, discounted to present value.   Money in is the real value of monero at the end of the project, when you plan to sell what you mined.  Money out is initial investment, plus expenses, less salvage value (when you're done mining and sell your equipment) discounted to present value.  Whichever project has the highest NPV wins.  

If you run the numbers this way, the open compute project comes out way ahead.  

There are two ways to compare an open compute project with a GPU project.  The first is to match the initial investment.  This would be a risk tolerant approach, where you leverage any appreciation.  Meaning if monero appreciates, you come out way way ahead, because you get way more hashes for your initial dollar.  The risk averse approach is to match hashrates, spending less upfront on the open compute project.  That way if monero crashes you haven't dug yourself into as deep a hole.  

This approach tends to keep your power expenses in perspective.  It's just one part of the equation.  

OP's points about intangibles are well taken.  I would add that the open compute project is scalable, if you can find a host for the right $ (they are out there).  GPU projects are only so scalable, you run into marginal cost issues as you cross the 15-20a barrier -- where you have to start figuring in the cost of renting and cooling the sort of place that won't ask questions about all those wires everywhere and all those sparks flying -- and the 200a barrier, which God knows what you have to do to make it fly.

Try xubuntu 14, should be able to install the proprietary radeon driver through the repositories.  Should get your hardware working, dunno about your mining software, what are you using?

Difficult discussion to follow without a clear understanding of what business yall are in.  Are you investing, turning capital into cash flow, or making an arbitrage play?

Say I have $20K sitting in the bank doing nothing. I want to buy a house a year from now and would like to make as big a down payment as possible.  Here are my options:

1.  Buy $20K in bonds for $19,400.  

I'll make $600, which is cool, but I lost $1800 in purchasing power due to inflation.  Which is not cool.  

2.  Buy gold.

I lose money in transaction fees and storage costs, but gold will tend to hold its real value over time.  Note, I don't really care what the exchange rate will be for gold a year from now.  If the price of gold doubles due to inflation, then I can expect the price of the house to double as well.    

3.  Buy a hot dog stand and work my fanny off.  

I figure I can make about $21K gross, $14K net, and can sell the stand a year from now for $10K.  I come out $4K ahead nominally, and a bit more than $2K net of inflation (until the IRS catches up with me).  

But I don't want a hot dog stand, I want an XMR mine.  And why not?  The numbers are the same, and I get to stay indoors and play xbox all day.  

Here's why.  My problem is, I can buy USD at a discount from a bond broker, and I can buy hot dogs at a discount from a distributor, but I can't get XMR at a discount.  So in other words, currency doesn't have a coupon. (Neither does gold.)

I can add value to the hotdogs by cooking them.  You can't add value to XMR.  Theoretically you could go down to Publix and  pay retail for hot dogs; they aren't worth as much raw as cooked.  You can't pay retail for XMR and expect to make money.  (Note: the definition of "exchange rate" is "retail price.")  

The similarity between the hot dog stand and the cryptocurrency mine is you're hoping to make money by moving things in time and space.  You can move hotdogs in space, from a distributer to a convenient sidewalk location.  You move XMR in time, from now, when it's illiquid, to the future, where we expect it to be quite a bit more liquid.  People will pay for convenience, or liquidity.  

"Arbitrage" refers to making money by moving things in time or space.  Speculating in currencies is neither investing, nor is it working.  It's making arbitrage plays.

You could say that by mining you are adding value, by digging that gold out of the ground, or by hashing transactions.  Fair enough.  But it doesn't follow that it's always a simple matter of dollars in minus dollars out.  

For a US citizen, mining, as it turns out, is a nifty sort of arbitrage "double play" on the Yuan.  Chinese want USD, and they are willing to give us computer equipment at a substantial discount to get their hands on some.  Interesting thing is, I think in a couple of years they are going to want to get their hands on XMR, to get around capital controls.  If I buy the mine at a discount, and sell the XMR at a premium, I make money coming and going.  Well, maybe.   Point being, that arbitrage is typically only a small part of turning capital into cash flow; adding value is the name of the game.  It's a huge part of currency trading.  

I don't mean to harp on XMR.  I am long XMR, but the same arguments apply to any cryptocurrency.  You gotta buy low AND sell high AND figure out where the arbitrage opportunities are, without getting confused by exchange rates and nominal prices.  The question is always --

1.  When to acquire coin.  What does the supply (emission) curve look like in light of present demand?
2. How to acquire coin.  Mine or buy?  If you buy, with what?  USD?  JPY? ZWR? (Don't say "ZEC" cuz I'll ask how you acquired the ZEC, and at what cost in terms of transaction fees and exchange rate risk)
2.  What is your exit strategy.  Who is gonna buy it from you, why, and when?  What does the supply (emission) curve look like in the future, in light of developing demand?

Right now, XMR is inflationary.  Meaning, the supply is increasing rapidly.   That will tend to keep its value low.  

The traditional way of looking at things is, you don't want to be long an inflationary currency.  You're better off borrowing.  In an inflationary environment, you borrow money to buy a house, with the idea being, your payments go down over time in real terms.  (How much stuff can you buy with $1200 today?  What about 20 years from now?)  

Cryptocurrency is different.  We know it won't be inflationary forever; in fact we have a pretty good idea of when the inflection is going to happen.  We know that demand for cryptocurrency tends to be low at first; it's illiquid, nobody knows about it, it has to "catch on" before it's worth anything.  But as we've seen with BTC, the demand is out there, and we even have a road map to see it coming.  

We know that it's easier and less expensive to mine when it's in its inflationary phase.

And we know that inflation isn't always bad.  The gold example shows classic economic thinking; that inflation is a global process that has no effect on real prices.  But that's not true.  Inflation can be sector-specific; we've seen sector-specific inflation in securities, housing, and health care costs recently.  And we've discovered that there's a "first-in" advantage; it takes time for new money to percolate through the economy.  If you can get your hands on it before everybody else does, you can buy stuff with it.  If you're the last to get  your hands on it, you can't buy squat with it, because by then nominal prices have gone up to reflect the increased number of dollars (or XMR, or ZEC) chasing around a given number of goods and services.  

First-in effect is a good argument for mining ZEC.  You don't really care why it's inflationary, all you know is, it'll be a long time, maybe forever before it's in general circulation.  It was also a good argument for moving to California in 1848.  Gold was inflationary but all it took was a pick-axe and a donkey and you could be first-in.  

I definitely think the answer to the  "when" question is, during the inflationary phase.  But my theory on XMR is that it's not really like mining gold, and you don't really need to bank on the first-in effect.  To me, mining XMR is like mining lithium.  Whether it's a good idea to go through the trouble depends on whether you think Tesla is going to take off.  If you think half the cars on the road are going to be dragging around fifty pounds of lithium in a couple of years, you don't really give a rats ass what they are charging you for electricity.  Your job is to fill that warehouse ASAP, and you kind of want to keep the market price low while you're filling it to discourage other miners, keep a lid on supply until you're ready to sell.  

The interesting thing is, an efficient market will take that expectation into account.   In a commodity market, where there is no expectation of future profit, there is little threat from new entrants.   If there is an expectation of great profits at some point in the future, competition can afford to drive short-term cash flow well into negative territory.  Profit expectation varies with time, certainty, and magnitude.  The more certain the outcome, the greater the future cash flow, and the sooner it starts coming in, the more of a short-term loss well-capitalized entrants can tolerate.  

This is what I find remarkable about XMR.  If you're mining say PASC, you probably don't care about future demand, you're probably flipping it as soon as you mine it.  If you're mining and holding XMR, theoretically you should be willing to take a loss now to acquire it.  The fact you can right now today get permission from your wife to mine and hold says some or all of these things are true:

1.  Barrier to entry is high.  People would get in if they could, but they can't.   And I think it is.  It only seems like everybody and their grandma is mining right now.   It ain't that hard to load claymore on your game computer, but not every dope can set up and run 20 or 40 or 100 gpu clusters.  There are people doing research in GPU computing who are trying to figure that stuff out right now.  It's not just a matter of learning facts already in evidence, or buying someone who knows them.  It's a matter of being way out on the bleeding edge of parallel computing. (For now, anyway.)
2.  Uncertainty levels are high.  And they are, but it's at least possible you guys understand stuff your average MBA doesn't have access to, right?  It's not just the ease with which you can slap together a rack of computers and make em work.  It's also the ability to see who the winners are, either because the math side of your brain can see where one currency will fail and another will succeed, or because you have enough "street smarts" or cross-competencies to see the leading indicators.  
3.  Demand is rising faster than supply.  I have a decent background in economics, and a fair amount of "street smarts."  Don't know squat about computers or cryptography.  But just looking at this thing strictly from a technical viewpoint, that's the one relevant fact in this entire discussion.  It's not that some people seem to want cryptocurrency.  It's that there's a humungous demand out there that's being partially eclipsed by inflation.  Go back and look at the emission/price relationship for BTC and you'll see what I'm talking about.  This demand exists in spite of entry barriers and uncertainty, both of which affect customers just like they affect miners.  I find that remarkable.  

I think I'm wrong about the lithium thing.   We are in the gold rush, brothers.  

I agree with OP.  We are living in remarkable times.  

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

(ETA: while I was writing this, NerdRalph posted a fantastic example of arbitrage.  Whether he made money or not depends on whether he held or dumped his BTC.  I would simply assert that it was good that he was just barely keeping his head above water while he was mining, because of *when* he was mining.  It's OK to break even or lose money if you're getting BTC at a discount because nobody knows what it is, or how to use it.  If you can sell it when the debit cards come out, and everybody wants some, you bought low and sold high.  Mining BTC right now is an example of the "greater fool hypothesis," where you buy high and figure some dope will pay an even higher price shortly.  Ralph did it right on so many levels.  That's where XMR is right now.  Only question in my mind is, what I'm gonna do next year, if anything.  Got my eye on them cannabis coins, they just need a non-tokin clear-headed MBA -- and a mathematician, and a programmer -- to lead them in the right direction.  Yo Ralph, I got the MBA, and if my kid ever graduates I'll have a mathematician.  If I can scare up a million bucks, are you in?)  

polaris bios editor.  which works on the elpida 470 but not the 480.  every once in a while i get a dead card, no system crashes at all.  part of the reason I run from console, it's the desktop and display manager that make things complicated.

\

@galaxySuser: I'm pretty new to the game too, so I have more empathy than knowledge but perhaps that will be helpful.

The ADL issues are discussed at length here and on github, there's a lot to wrap one's head around; here are the basic issues as I understand them --

- When you're compiling, you have to copy the libraries into .../sgminer-gm/ADL_SDK (there's a readme file in there that explains)
- Personally I haven't found that the temp monitoring works from console (either under AMDGPU-PRO or fglrx)
- Temp monitoring works just fine for me in the graphical environment running fglrx, but there have been issues with AMDGPU-PRO, for example ADL and OpenCL sometimes disagree about which graphics card is which.  

The developers are working on it, and there are solutions (scroll up a bit) but keep in mind there are a mind-boggling number of moving parts here, mostly coming from the fact that Radeon woke up one morning about three months ago and found out their drivers were incompatible with pretty much every linux distro in the world.  Right about the time the RX 400 series hit the market. Legacy support is really good right now, but 400 series support is still a work in progress.

For a casual hobbyist, there's an argument I think for 2nd or 3rd gen cards running under xubuntu 14.04 and fglrx, closest thing to hitting the easy button there is.  With the 400 series, good news is, they hash OK with low power use right out of the box.  Bad news is, to squeeze the most out of them, you have to bring anywhere from some, to a heck of a lot of expertise to the table depending on what equipment you're using.  

Me, I bought a bunch of fans.  Seems to be working.


@citronik:  hey, not bad.  I am surprised.  Still haven't cracked 600.  Getting close tho.  I get the feeling it's about architecture, quality of components, and memory timing.  Not sure I'm gonna make much more headway until I optimize the timing.  Clock speed seems like kind of a blunt instrument.

Looking for advice tuning my 480's for XMR.

Rig:
-Gigabyte ga-970a, phenom IIx4, 4g RAM.  
-Xubuntu 16.04LTS (running from console), AMDGPU-PRO 16.60
-4x sapphire rx 480 nitro 4gb elpida, stock bios.  Stock clocks are 1143/1750

Hashrate is hovering in the mid 500's.

Settings:
- w8 g1 (haven't seen much difference going with w4 g2)
- ri currently 1088 (started at 1152, not much difference backing down as low as 1024)
- I've tried adjusting the gpu clock between 800 and 1300 and the memclock as high as 2200, and can't say any of those change a durn thing.

Had the same experience with my sapphire 470/4gb (elpida) cards.  I could occasionally get em up to 600 in windows by flogging the memclock, never got above 550 in linux until I did a memory strap (1500 copied up), and now I'm getting solid 600 on stock settings.  They run pretty hot but I think that's due to crappy reference design fan.

Not eager to flash the ROM on the 480's.  No bios switch.  Polaris bios editor doesn't work with these cards. Tempted to just live with it, in lieu of downloading a bios off the internet.  I'm open to suggestion.

By the way, these cards are on sale at Newegg for ~180.  Net of everything it was probably a decent deal, it was worth the extra 10 bucks over a 470 to get some decent fans on there.  If I can squeeze a little more out of them, it'll be a great deal.  

Package landed and everything is in great shape.  Excellent communication, shipped fast, everything was packed very nicely.  If you're new to mining or thinking about getting into it, these are turn-key packages for a great price. Jump on it.

Rocky, thanks.  We appreciate your service.

For XMR:

Review results on the "monero benchmarks" site, just remember that flogging the card is not the point.  It's like tuning a motorcycle, it needs to be right for the race you're running.  This race is about fast access to slow memory, basically.  Architecture-wise, Gen2/Hawaii got it right.  Main advantage of Gen4/Polaris is energy efficiency. By the same token, Gen1 cards are working with a handicap, that can be compounded by crappy memory and crappy timings.  Comparing 7950/280x with rx470, on average I get probably 60% of the hashrate and burn I dunno ~60% more energy per card.  The gigabyte ime is the biggest piece of crap ever, I'm gonna keep the cooling system, which might come in handy someday. And think of some way to blow up the board.  Might launch it into orbit with a cluster of bottle rockets, that would be cool.  I'm no fan of that Sapphire 280x; crappy memory, cheap card, not much you can do to tune it.  

With regard to your Gen2/Gen3 cards, if you're following a "mine-and-hold" strategy, you're banking on future demand.  The profit equation has to include time value of money as well as energy costs.  Given prevailing interest rates, time is on your side. (Well, that's a strong statement; let's say it's not exactly not on your side.)  If you think XMR is going to pop, you can absorb some energy costs now.  

Conclusion:  Keep the 290's and the 390 for XMR.  Hashrate will crush, don't look at the electric bill.

 

For Dagger-Hashimoto variants:

I'm sure you know more about ethereum-style algo's than I do, all I can say is, my gen-1 cards are doing well with PASC, which is in the same general ballpark I think.  I've been shying away from ETH, because of concerns that the difficulty is going up, although maybe I should give my 3gb cards a chance.  With PASC being newer and more of an unknown I'm assuming I'm more likely to come out ahead.  I have no idea what the deal is with PASC and don't care, I mine to an exchange and flip it.  

Conclusion:

Me, I'd dump the Gigabytes if you can.
I'm not sure what to do with the Sapphire, might give it a chance with PASC.  But I think it has some resale value and not sure I want to flog it too hard.
Personally I'd hang on the Asus's, get multi-algo software, mine to an exchange.  Take what you can get and flip it.  

Ebay:

Seems like the Gen1 cards -- at least the ones that look spiffy -- go for anywhere from $80 to $100 on ebay.  Gen2's sometimes go as low as $90.  

Gen4's are going for as little as $160 (check NewEgg, which takes BTC like it was paypal).  Avoid 460's.  I'd favor Sapphire.  Their "platinum OC" 4gb is a fine card for the money; I guess it must be better for mining than gaming.   Looks like the nitro's are faster, although I'm not convinced the 8gb nitro 480 is worth the money.  Probably has the best resale value, and it's a "keeper" for games.  But it's not clear to me that 8gb gives you much of an advantage over 4gb for monero, and the marginal cost of the extra compute units is exorbitant, $20 vs $5. 

Gen2's are faster than the Sapphire references, on my rigs I'd say 3 Gen2 vs 4 rx470's is about a wash all the way around for a hobbyist.  Little doubt in my mind that the 400's will have more salvage value.

Conclusion:

I'd dump something, and given the prices on 400's, I'd try to acquire one or two to play with.  

does flashing the bios help those cards?  elpida ram usually comes pretty well-timed right out of the box, doesn't it?  it helped mine some, I guess, to the extent I get the same results with memclock at 1700 as I did before at 2000.  can't see it does a dang thing to the voltage.  not a ton of info out there on the topic, mostly talking about hynix seems like

I dunno.  Far as XMR is concerned, my feeling so far is Gen1 is Gen1.  Got a rig with 4x7850 hasing 1.2kh/s drawing like 430 watts or so.  Any combo of three 7950/R200 is about the same.  

I get the feeling mining XMR is all about giving the compute unit good access to good memory with good timing.  According to the bitcoin wiki, cryptonight was originally a CPU protocol, and the defense against ASICS is dependency on rapid access to an amount of L3 cache that isn't practical for an ASIC.  GPU mining then is a handicap, the memory is slower and the compute units can't get at it as quickly. Going from 7850 to R9280 makes more noise but I don't see that it's doing any more work.  Granted, the 7850's have decent RAM, and elpida ram seems like it comes timed right from the factory most of the time.  And the 200's are spectacularly crappy in the memory quality department.  Boils down to paying more money for lousier RAM, basically.  

But going from a 280 to a 290 -- lousy RAM and all -- now we're talking whole nother ballgame.  My 290x smokes everything in my collection, including my 470's.  AMD said they were giving the compute units better access to memory, and they delivered imo.  

Only reason my Gen1 200's haven't been drop-kicked into the woods is because they hash Pascal like it's going out of style.  Still trying to wrap my head around that one (I mean, my kid is a senior math major, I'm like when and at what cost are they gonna start teaching you cryptography already) but I think it's one of those ethereum things, where the total amount of RAM at your disposal is a big factor.  

Only reason I haven't crowned the 290 the el cheapo king is there's too many plugs on it, and I can't figure out how to hook em up to my el cheapo HP server PSU's.  Also it's hard to cram it into a milk crate.  

Interesting stuff in the database.  Interesting lack of correlation between specs, price and hashrate.  Also interesting to see what the fancy CPU's are up to.  A Dell workstation motherboard with five PCIe slots, two E5 processors, ram and heatsinks is about $400, about what the windows folks pay for a new mb with one chip.  Wonder if that kind of hybrid is the way to go.  I'd gain about as much as adding another graphics card, albeit at a cost of probably 300w.  If I could undervolt that sucka....I dunno. Might be worth doing just to say I did it. I could put a rack of 290's on there and blow the transformer. That would be cool.

Interesting.  Haven't figured out how to compile sgminer without the graphical environment, but once it's built, I can run it from console.  Most of the time.  

Been running sgminer-gm from console in ubuntu 14.04 w/fglrx for a while now.  No issues whether running from host machine or ssh.   Ubuntu 16.04 is kind of an enigma though.  Systemd is a horse of a different color.  Right now I'm using a suspenders + belt strategy, have graphical.target set as default, disabled lightdm.service, and backed it up by with grub_cmdline_linux=3 and -default=sytemd.unit=multi-user.target.  It's hit-or-miss.  Sometimes I can get it running, sometimes sgminer can't find my cards.  Oddly it seems to run better from the host machine than ssh, although I haven't been able to consistently reproduce that.  

Looking at my fglrx rig, everything seems more stable and easy to work with when I kick X to the curb.  No more hard crashes.  No crashes at all if I leave it alone.   If I go ahead with plans to put in 10 or 20 rigs, that's the way to go I think.  But, if I do put in a bunch of rigs, it'll be 470's.  So I guess I gotta figure systemd out at some point.  Or else kick that to the curb too, which is part of the appeal of Arch I guess.  I mean, a lot of the stuff that gives us trouble (display managers, desktops, etc, and I include systemd in that) is nice running on your laptop, but doesn't really have a place in a dedicated mining rig, and just interferes with the ability to set it up how you want.  That's the feeling I'm getting anyway, so far.

Anyway sgminer-gm is tight.  Be glad when everything else settles down.  

Awesome, on so many levels.  I only have two suggestions --
1.  Put in a bay to hold a couple of server power supplies to drive the PSU's.  (Which, I just spend the last hour looking at psu backplanes thinking, I could rewire that...)
2.  Paint it black and up your price to about $200 grand or so.  I mean, watch the video on the "Klimax 250" which I am not making up.  

Yep.  That's pretty awesome actually.  Guess it's linux-based.  Hard to tell.  Hard to blame him for soft-peddling that, I guess.  

Wonder how much graphics simplemining is pushing through the pipes.  Wonder if it needs X to be running on the host machine.  Just thinking with the MBA side of my brain, I don't think that's the optimal way to do it. I don't think that's the direction it's going.  I see the AMD world tossing X to the curb. No idea what nvidia is up to, but I think OpenCl is the eye of the storm, and from what I've read, that is and has been moving in the direction of bypassing X altogether.  Not quite there yet, I don't think.  But I think that's where it's going.

I'm assuming there's always been client-side software in the mix.  Question I have is, can ALL the graphical stuff be client-side.  I mean, it's not 1984 anymore, right?  

Got a nickel says your average miner would do anything to ditch the graphical environment.  Suspect most do, if they can.  Suspect those who do bring a lot of expertise to the table, or else pay for it.

So yeah, could be an opportunity there.  Host-side product that is delivered (and maintained) with drivers, software, scripts, networking and security baked into the pie.  Client side software ported to...  well that's the thing.  Kinda feel OP is looking at it upside down.  Big task trying to figure out how to build from scratch.  Build the app first, and I think everything else will start to look quite clear.  Put it this way.  When he starts looking to see what's lurking beneath the surface of android, chrome OS, chrome browser, steam --  I think he's going to be pleasantly surprised.

Now, the question is, where does he add value.

I'll start with the proposition that miners aren't precisely "making money," and speculators aren't precisely "investing."  We are engaged in the process of providing liquidity for people who are willing to pay for it.  That's what gold miners do, and that's what we are doing too, only better.  No gold miner ever offered to guard your gold against theft, keep your books for you, and open up a "Paypal Gold" account for you.  OP is right about that, and right about how important decentralization is to the whole scheme.  

And he's right about the need to get power to the people, because up until now, liquidity was a government problem.  Your average guy seldom runs into a liquidity issue.  I mean, if you really have to sell your car right now today to afford tuition, yes you have a liquidity problem.  If you don't have enough assets to pay your tuition, that's different.  

Things are different now.  Every man, woman and child in India has a liquidity problem right now today.  Every Chinese dude who wants to invest in American or European securities has a liquidity problem.  Every guy in Venezuela or Zimbabwe who is tired of working for "money" because he can't convince the guy at walmart to trade toilet paper for it has a liquidity problem.  

This is the interesting and subtle part about OP's proposal.  It's that jillions of ordinary folks around the world are gonna be willing to pay for liquidity (and anonymity, and hedge value), and that it would be kinda good for ordinary folks to facilitate that.  I mean, we don't want to double the hashrate overnight.  But we don't want the network to be dominated by a few government-approved whales, either.  

Where he's not quite on it is the ease-of-use thing.  Screw that.  That misses the point entirely.  It's not that, I dunno, Africans are too stupid to figure out network security.  It's that everybody's time is worth something, and if people are going to run a node -- which is what it's really all about -- it needs to be worth their time.  The only entity that can consistently afford to waste money is the government, and that's how the whales get in.  

It's about making it work.   It's about making running a node businesslike and efficient.  And I think the key to the whole thing is getting rid of unnecessary complexity.  Which means, getting rid of unnecessary overhead and uncertainty.  

Look.  I like my Mac as much as the next guy, but once I got the idea in my head to build 10 or 20 mining rigs, it occurred to me that the GUI is a Faustian bargain.  I'd rather eat worms than manage a 20-node Windows network, and Ubuntu isn't much better.  Either way, I'm always one upgrade away from down time.  In both cases, I think the graphical environment is the problem.  Of course it is.  I look at my work computer.  It's not just that they make us use Windows, not just that, as a result, they have insane hardware and IT costs.  It's that they make us use Citrix.  Which means, we are running Windows inside Windows.  Even when it works, we are trying to cram all those graphics through these little tiny pipes, which wouldn't seem so tiny except to the extent that we make them so.  I can't believe they can afford to pay me what they do to sit there and watch the spinning hourglass of death every time I click the mouse, not to mention having to boot Windows multiple times every day, because it never works for long.  Not to mention having to boot it TWICE multiple times every day.  Who thought that was a good idea?  Have we lost our minds?  And running xorg-xserver-GDM-metacity-unity on 20 machines that never interact with a human is even worse.  I mean, windows can't help what it is.  Linux should know better.  

Meanwhile I've been trying for the last month to prototype a rig, and every time I turn around I'm changing my protocol because some new version of radeon-turbo-nitro-OC-pro won't talk to this week's version of X-almighty-whatever, not to mention the fact that my motherboard is constantly accusing me of feeding it the Stuxnet virus.  And just to be clear, the kernel is sitting over there saying "I didn't do it," and I believe it.   It's the graphical environment every doggone time.  If I could figure out how to compile the software without it, I would.  Fortunately, it doesn't need anything with an X in it to run, so I kick that sucker, and the trash it hangs out with,  to the curb as fast as I can.  My feeling is that the people writing the code feel much the same way.  

Now, I'll startxfce4 (or nowadays I guess I have to ask systemctl to handle it, don't even get me started) if it's the most efficient way to do what I want to do, and sometimes it is.  It is not the most efficient way to keep my machines running and talking to each other.  It's the opposite of that.  

That's the point.  Ease of use ain't the thing.  I'm not stupid,  I'm busy.  So if you're talking about my money?  Deliver me a host product that works, and keeps working.  And a client that lays out what I need to know without having to ask for it, basically.  You're doggone right I'd pay for that.  You don't need to make it easy, not exactly.  I mean, we don't engineer the cockpit display in an F22 precisely to make it easy.  We lay it out in such a way that the human brain has direct access to critical information without going through a translation paradigm. We are removing clutter, removing intermediate steps, taking some of the workload off the brain.  Removing unnecessary complexity and overhead.   I know that's what you really meant, or at least on one level I think that's what you meant.  Stay focused on that.  

Key to the whole thing, OP, is to solve the right problem.  The GUI is the problem, not the solution.  We baked the icing into the cake, which was a mistake.  If you can dig us out of that mess, you win.  

They recommend copying the adl headers to the /sgminer-gm/AMD_SDL folder.  But they are still working on that part, doesn't work all that hot.  Don't need it to mine.  What kind of card do you have, and what's it hashing?