Alternate title "
The only way forward to real security is ASICs, but it is going to hurt in the short term".
Background/theory (skip if bored)So difficulty is going up a lot but I think (based on anecdotal evidence) on the forums I think many underestimate by how much. People see difficulty going up x% on average per adjustment and make statements like "well it can't keep doubling forever" and while that is true it misses the point. We don't know exactly how much difficulty will go up each 14 day period as it has more to do at this point with supply chain constraints than actual demand for new hardware. It is better to look at where hashing power and thus difficulty is going to and then work backwards from there.
Looking over the various pre-order threads (BFL, Avalon direct, Avalon chip assemblers, KNC, ASICMiner, Bitfury) conservatively I am taking a SWAG and saying we are looking at more than 5,000 TH/s of hardware ALREADY ordered* (see disclaimer below). This "preorder bottleneck" distorts the normal new hardware deployment feedback cycle. The feedback cycle is where new hardware gets purchased, deployed, hashing power and difficulty goes up, and that forces the return on new hardware down. This lower return (ROIC%) makes new sales less attractive and hashing power increases slow and eventually flatline. What usually leads to a resumption in higher difficulty is a higher exchange rate which raises the ROIC% in USD terms. My assumption is that we wouldn't be seeing 5,000 TH/s ordered at current exchange rates if it weren't for the pre-order bottleneck ("peB" from now on because I am lazy).
The peB allows those ordering hardware to delude themselves. While logically we know difficulty is going up, we don't know exactly how much or how fast, or who will deliver first. This leads to a scenario where an individual prospective miner can fall for the "I am special" fallacy. In a casino most people go in knowing that "most people" lose money in a casino (those suckers), yet every gambler believes they are "special" and somehow have a reduced chance of being in the "most people". They reality is they don't but when down, they just need a run of lucky cards, and when up they can't lose. Inevitably in the long run they are no more special than anyone else and join the ranks of "most" losing their money to the house. This insight only occurs after the fact. Without the peB, if difficulty today were 500 million, the daily return on 1 GH/s of brand new hardware would be ~0.03 BTC per day. That would make prospective miners less likely to deploy new hardware. Although it is highly likely difficulty will be higher than 500 mill by the time most miners get their rigs many people believe they will be the lucky first and thus while many will lose, they will make enough return fast enough to beat the rising difficulty. The reality is like the lucky gambler a small few will but most won't but this results in more hashing power being pre-ordered then would be ordered.
Eventually these 5,0000 TH/s of preorders will be delivered and difficulty will rise to ~800M. How long will it take? 6 months? 12 months? I don't know but it overall it doesn't really matter. See the hashing power increase is directly related to the speed of delivery. Slower delivery, means slower growth but most miners will not have rigs yet. Faster delivery means more miners get their efficient rigs but difficulty will rise faster. When someone is hoping they get their rig "fast" what they really mean is I hope I get my rigs faster than the average guy but the majority of people can't get their rigs faster than the average guy. Due to this most miners will get their rigs "too late" for higher profits and it doesn't matter if "too late" is next week or next year.
Another way to look at it. Among the various ASIC offerings the average price is around 40 MH per dollar. Lets reverse that metric, one GH/s costs $25, or one TH/s costs $25,000. If 5,000 TH/s have been ordered we are looking at $125M in pre-orders. Electrical efficiency varies a lot but if we use the median of ~300 MH/J the entire network will produce a load of 16,667 KW (5TH/s / 300 MH/J / 1000 J/s per KW = 16,667 KW). Avalon/ASICMiner will have double the electrical cost and BitFury will have half the electrical cost. Also once again the exact numbers don't really matter. We are just looking at the magnitude of the values (so if it is an average of 400MH/J or $30,000 per TH/s it doesn't really matter). You can run with marginally different numbers they really don't materially change anything. What would change things is a magnitude difference (i.e. 3 GH/J or $2,000 per TH/s) but those are highly unlikely in the near term. With 8760 hours per year (24*365) and $0.10 average electrical cost, the network (5000TH/s @ ~300 MH/J) will expend ~$15M in electrical cost electrical costs annually. Once again how profitable each individual miner will be depends on timing, efficiency, longevity, and yes a lot of luck but we are more interested in all miners collectively. As a group all miners share a "reward pool" of ~$123 mil at current exchange rates (6 blocks per hours * 24 hours per day * 365 days per year * 25 BTC per block * ~$94 USD per BTC). $123 mil - $15 mil in electrical cost = $108 mil net reward on $125 mil total deployed capital assumming no new hardware (or change in exchange rate).
The problem is 420 days is forever and as the backlog diminishes demand for new hardware will go up. Maybe miners are smart and won't accept the current prices but ASIC producers have plenty of margin and can cut prices 50% or more to stimulate new sales. It would be foolish to do so now with massive backorders but eventually those will clear out and those producers aren't going to just close up shop. If $25,000 per TH doesn't cut it maybe $15,000 per TH will or $10,000 or $9,5000. Due to that many miners will see their 420 avg break even being continually pushed back into infinity. Sorry.
So what will things look like when pre-orders are delivered in bulk.
Hashing power ~5,000 TH/s
Difficulty = ~800 million
Revenue per GH/s: ~0.01886 BTC per month
To factor the average increase per month (or day) just estimate how long it will take to delivery this hardware.
Will it take a year from today?
(5000 - 200)/ 200 = 2,400% or ~0.9% per day compounded growth (1.009^365).
Six months?
(5000 - 200)/ 180 = 2,400% or ~1.8% per day compounded growth (1.018^365)
How high can difficulty goWell once paid for hardware is a sunk cost. An individual miner can sell a rig to another miner but that doesn't change global hashing power of difficulty, it only changes who is taking the risk. Also unlike GPU where they could be sold for non-mining purposes and thus remove global hashing power, ASICs only have one purpose. So unless miners in bulk decide to re-enact the copier scene from Office Space the hashing power will still exist.
How high depends on how low of a return miners are willing to take. Since miners (collevtively) do a bad job of estimating what other future miners will do (and thus raise difficulty) we can guesstimate that difficulty will aproach (but likely not reach) a point where the electrical operating cost of existing hardware equals the mining revenue, the "break even on electricity" point. If the average rig has an electrical efficiency of 300 MH/J, the average miner pays $0.10 per kWh and the exchange rate remains the same that break even (excluding capital costs) point would be:
Total annual revenue: $123 mil (6 blocks per hours * 24 hours per day * 365 days per year * 25 BTC per block * ~$94 USD per BTC)
Total electrical consumption (kWh): $123 mill / $0.10 per kWh = 1,230 million kWh (or 1.2 trillion kWh)
Total electrical load (KW) 1,230,000,000 kWH / 24 /365 = ~140,000 KW
Total electrical load (J/s): 140,000 KW * 1000 J/s per KWh = 140,000,000 J/s
Total MH/s: 140,000,000 J/s * 300 MH/J = 42 billion MH/s or 42,000 TH/s
BTW 42,000 TH/s (or 42 PH/s)
Now this is the break even electrical point for the average miner and logically hardware isn't free so miners (collectively) should stop adding new hashing power well short of this however collectivley miners aren't always that logical. This could be considered the upper bound for difficulty in the ASIC era. Difficulty will not sustainably rise beyond that unless
a) exchange rate rises and remains significantly higher than $100. i.e. @ $200 the break even difficulty would double.
b) ASIC producers significantly lower their selling prices. While there are a lot of potential margin to cut. Maybe $20,000 per TH/s is possible, maybe even $10,000 per TH/s is possible.
c) ASIC producers are able to improve efficiency. With most designs at 40 to 55nm there isn't that much that can be cut here. A jump to 28nm might give a 2x increase, and 20nm (someday) another 2x increase (say 4x theoretical from today).
For b & c the "better they get" the less potential improvement exists. At best an ASIC producer could only sell at cost on a cutting edge manufacturing process (20nm today). Obviously there is no reason to do so but we can look at that as once again an upper bound on improvement. Baring some radical break through (i.e. a exploitable flaw in SHA which allows better than brute force mining) baring an exchange rate increase difficulty will remain within one magnitude (up or down) of 40 PH/s.
Wait 42PH/s that can't possible right? Can it?Lets look at it in reverse.
In January the network was a majority GPU miners (with a small amount of FPGA & CPU miners on the edges).
If you look at the difficulty & hashrate between 9/12 to 2/13 it was roughly (+/- 25%) 20MH for that period of time.
Note it isn't important to look for the exact hashrate but we can see there were no magnitude changes (i.e. 2M or 200 MH).
Here is a pricing chart for that period of time.
Volume weighted average price is ~$15.00. Now this year price rose from $15 to $266 and then back down to $90. I would argue that the network was in equilibrium from 9/12 to 02/13. Essentially the revenue for miners roughly balanced the risk and thus expected ROI%. The only thing which prevented a rise in difficulty due to more GPU rigs was the "future risk"of FGPA/ASICs. In another universe where FPGA and ASICs were impossible I would expect that when price rose by a factor of 10x and thus miner gross revenue rose 10x that hashing power (and difficulty) would have also rose 10x.
So all things equal a GPU only network would probably support a hashrate of 10x 20MH = 200MH at the current exchange rate. Once again the exact math isn't important but it is unlikely it would support 2000 MH/s or only 20 MH/s today (even with only GPUs).
So if current exchange rate can support a GPU powered network of 200 TH/s and the average ASIC is 150x the electrical efficiency (MH/W) and 40x the capital efficiency (MH/$) we would expect that ASIC miners (who probably are no better or worse at projecting profits) would drive hashrate up by at least 40x and probably more like 150x. Over time electrical costs become more important as people are more likely to overbuild the network then underbuild as the under predict the rising difficulty and once purchased hardware is a sunk cost.
40 x 200 TH/s = 8 PH/s
150 x 200 TH/s = 30 PH/s.
Remember 40 PH/s is the upper bound based on break even efficiency. Without a peB and high availabity of ASICs I wouldn't be surprised to see the network already at 8 PH/s on it way towards 30PH/s.
TL/DR version:
the network "only" being at 200 MH/s is a combination of
a) slow deployment of ASICs
b) lack of new GPU hashing power because of the risk of the inevitable future
One shouldn't look at 40 PH/s and say "that is 2000x higher than today with ASICs" as the 200 MH/s today is artificially "low" (although it may not feel like that for GPU miners) for the two reasons above.
Hey jackass, this sounds like horrible news, how is me losing money "ok"?Well there will be pain. The peB caused surplus capacity to be built and that is going to result in some losses by at least some miners. I don't wish any losses on anyone and if you are afraid of losses well you probably shouldn't order, seek a refund, or sell your spot in line. However this is good for Bitcoin in general. At 5 PH/s the cost to attack the network is probably greater than $100 mil and that is probably a lower bound on hashing power. Since electrical efficiency break even (for current gen & exchange rate) is closer to 40 PH/s, if the network eventually rises to 20 PH/s the cost is closer to a quarter billion. Could the US govt pay that? Yes but security isn't a black or white dynamic. It now puts 51% attacks well beyond the range of all but the most power sovereign entities. Even the largest corporation's can't just pretend away a 9 figure expenditure as if it is nothing. Whole ranges of attacks are now gone. An economical 51% (where attacker attempts to profit directly from the double spend) is a non-issue, botnets will never have the capability to threaten Bitcoin. Hell Bitcoin is even safe from the majority of Sovreign nations. From genesis block to a quarter billion dollars in security in 5ish years. Not too shabby.
The other piece of good news is that even if miners will not profit it makes no economic sense to stop mining once the hardware has been paid for. The overcapacity due to pre-orders may result in reduced future hardware sales but for existing miners, their hardware costs are already sunk. Any mining revenue means a smaller loss (and potentially the hope of a small return). Turning off one's already bought miner is just a way to guarantee a maximum loss. Think about it assuming revenue is greater than electrical cost at which point does a miner realize the maximum loss (assuming no resale)? The maximum loss occurs upon delivery if miners, doesn't mine anything. Even if the miner will eventually come out behind the longer the miner mines (when revenue is greater than electrical cost) the more net revenue against the same fixed capital cost. Miners who in frustration sell their rigs (because they realize the break even point is 1,000+ days away) to other miners don't really change the security of the network.
This level of security is simply not possible without ASICs. ASIC "unfriendly" designs simply pigeonhole themselves into a space of minimal security. Maybe good for a hobbyist but hardly strong enough for any real commerce. The good news is that with Scrypt's memory hardness reduced by >99% only existing Scrypt based coins are only marginally ASIC resistant.
* Disclaimer: try not to get hung up on the exactness of 5,000 TH/s. If you think it is off then substitute your numbers instead. The reality is if only 1 PH/s of hardware is sold then more will be sold once pre-orders are delivered until the market reaches equilibrium. Really it only changes the shape of the curve not the endpoint.
