Topic: Impact of ASIC on BTC price - Lowering the Marginal Cost of Production (Read 5194 times)

Agreed with those statements.


But this wont happen until ASICs are priced near marginal costs.  We will almost certainly start several orders of magnitude above that. Its the inbetween, that is the problem, at least for miners. And that inbetween could last pretty much forever, because its highly likely most asic miners will never make a positive ROI as prices drop and difficulty goes up in a race to the bottom; there is no point in turning off those asics either as long as they produce more than the electricity cost. IOW, I expect the market will get saturated even before prices have reached marginal costs. As a result, every asic customer will lose money, not operationally, but when you factor in their investment. So unless BTC exchange rates make some large and unforeseen jumps, profitable mining will be a thing of the past.

I'm trying to understand this and for me the arguments about what BFL, or any other ASIC maker, may or may not do is not important.  If you make an ASIC you will price it so as to maximize profit, and as long as there is profit there will be competition.  If the ROI for ASIC hashing devices is too low, then they will not sell and manufacturers will lower the price or go out of business if they are already selling at the lowest marginal price.  

Once ASIC devices, at the lowest cost to manufacture and highest efficiency possible are not economically feasible at current difficulty levels, the only thing that can make them feasible is an increase in the demand for BTC, difficulty levels dropping due to miners dropping out and/or miners willing to accept a lower ROI.  I would like to examine the ROI aspect.

As the risk of difficulty increases becomes more stable, miners should be willing to accept a lower ROI commensurate with the decreased risk to their investment; difficulty should still increase at some equilibrium rate but will be bounded.

As the demand for BTC becomes more stable, miners should also be willing to accept a lower ROI commensurate with the decreased risk of the investment due to currency exchange fluctuations.  Increasing BTC prices will also drive difficulty to increase.

So in general, difficulty will increase and ROI rate will decrease when the rate of difficulty increase is stable.  The primary factor governing the ROI rate will become demand for BTC.  When the BTC outlook is positive, it will make sense to increase investment in mining, and when the outlook is uncertain, demand for mining hardware will decrease.

Taking everything out to the extreme, at some point the ROI rate for new mining hardware will reach the lowest ROI rate people are willing to accept given the intrinsic risk from a steadily increasing difficulty rate and uncertain BTC demand (exchange risk).

Eventually demand for new mining hardware will disappear unless there is a stable increases in BTC demand.  Manufacturers of mining ASIC hardware that have paid off their non-recurring costs will probably stay in business but their fortune will be tied directly to increases in BTC demand.  Likewise, people who have invested in mining devices will only be able to justify increasing their investment when the outlook for BTC demand is increasing.

Perhaps someone with more analytic training can solve this equation: for some fixed estimated price per performance of an ASIC device (eg. $50 for 1GH/s) at what difficulty, difficulty increase rate and/or BTC dollar price can an average yearly return on investment of 20% be achieved.

For example, a device today priced at $600 for 800 Mh/sec, with a current difficulty of less than 2Mh and a BTC price of $5/BTC will yields at least .4 BTC per day, or around $2 per day.  If that could be achieved consistently it would yield around $700/year.  

The only way that this is not a fantastic investment is if difficulty increases (as is likely) or the BTC price falls (which is possible). You won't find 100% average yearly returns anywhere on wall street.  But adjusting for the risk, this is a well justified return.   What looks like 100% yearly ROI today, could easily become closer to 0% tomorrow.

But that said, as difficulty increases become more stable and predictable, the return on their investment a miner is willing to accept should become less also.

If ASIC mining devices lead to a decrease in volatility in difficulty increase, mining hardware will continue to sell, and difficulty will continue to rise until the point when any further increase in difficulty will decrease predicted returns below what you can get from a safer investment of your money.

its not cost that will determine price initially, its demand. At current difficulty, demand will enable BFL insanely high gross margins (which are needed to pay back the NRE), easily 10000% or more of marginal cost;  but as difficulty increases, prices will come down accordingly, or miners will no longer buy them; eventually its the marginal cost that will determine prices.

lower prices -> increasing difficulty -> causing lower prices; its a feedback loop you do not have with cpus, gpus or fgpgas because their price is not determined by bitcoin difficulty. Thats the key issue, the difference between initial prices and eventual prices per GH will likely be several orders of magnitude, since no one is going to buy asics that cant generate a profit in their lifetimes, and there is no reason BFL wouldnt want to sell products as long as they still have a huge gross margin.


See the many other threads on this issue; its based on some academic research and an actual SHA256 asic implementation.

But the fact of the matter is that even if every single bitcoin miner bought an ASIC to mine with, the mask costs are still a significant part of the overall cost of bringing one model of ASIC to market. Thus all these models that assume negligible marginal cost of production results in a very low equilibrium price are suspect.

How do you estimate this number?

Mask set is an NRE, no one ever said NRE was low, quite the opposite (and a few 100K is if anything, too low an estimate). We are talking variable production costs being negligible.  A fully processed 200mm wafer costs around $1000 for an older 130nm process. One such wafer should yield somewhere between 1 and 10 terrahash. $0.1-$1 per GH is negligible compared to current prices and asic development costs.

ASICs aren't negligible cost to produce. A good mask set costs a couple of hundred grand and a good rough estimate for cost for a 300mm wafer is $5000.

BFL would literally need to sell thousands of these at hundreds of dollars to break even.

I'm curious on how they plan to do so which makes me suspicious of the ASIC announcement.

Why do you imagine ASICs are so exclusive... sure the first run will be quite expensive but prices generally come down rapidly as volume goes up.  At that point anybody will be able to buy one and mine at about the same cost/MH as the pros, if connected via USB to a netbook or something that sips power.

The fact that you have to have a particular GPU to mine profitably (and really you want to pack as many as possible into a single machine to avoid electricity overhead) already puts mining in the hands of the pros.

The best argument for general ASIC availability is probably for the security of bitcoin...if the market moves to the fastest/most efficient solution around, it will make it much harder for a large entity to grab 51%... with the intent of shutting down the network by not allowing anyone's txns for example.

Even switching to an algorithm that favors CPUs over GPUs is a bit dangerous, due to the existence of special-purpose massively cored CPUs.  For example this one's got 100 cores: http://www.tilera.com/

What happened to the idea of regularly modifying the algorithm so that ASICs are not worth it?

Even that wont help; you are suggesting they keep the price high to prevent difficulty from going up too fast, but a high per GH price also linearly increases the payback period, so the period BFL would have to limit sales to keep difficulty more or less stable. Thats not gonna happen,; BFL cant keep selling machines at the same price if difficulty goes up and BTC value does not.  BFL will have to lower their price or sell not nearly enough to make up for their investment. More so if they worry about a competitor, which they probably should. 

So those people who buy them early at maximum per MH price are likely shafted just like the ones that will come after them, particularly if there are many of them, which I think there will be; at almost any price point I suspect there will be a race to get those machines early from people who just dont realize whats about to happen.

The only thing BFL can do is limit sales; not even by raising the price, but simply limit the number of GH they sell even if there is a big demand for their product. And when you think that through, it means having money printing machines on the shelves that you dont monetize. In short, BFL will have the choice between screwing themselves or screwing miners.

I'm just guessing here but FPGA's are going to improve over time also, especially with the incoming 22nm process, effectively doubling hashes, and ever better mining firmware (another area where ASICs are lacking, they can't be improved without re-manufacture). Even GPU manufacturer, a place where there is vast investment in research and design, might play a part far into the future. As an aside a lot of these FPGA boards can easily be re-tasked since they can be flashed so re-sale my not be as bad as predicted which can't be said for ASICs. Every one of these hardware solutions is going to be prone to Moores curve so even GPUs will likely hang on far longer especially for those with marginal electricity costs. I think it's too early to call but predict that FPGAs will have a very strong foothold by the time ASICs become a real issue.

My point is that even if Bitcoin is a niche what's good for it is actually good for the industry at large so any improvements in GPU, FPGA and ASICs are likely to be beneficial to Bitcoin in the long run.

This is exactly why BFL et all should price their products to only have a slightly better ROI than GPU's.  Otherwise they simply fuck over their own market. 

As others here have stated, eventually ROI for Bitcoin mining will approach infinity and be owned by individuals who can never reasonably expect to make their original money back. Perhaps later, when they petition the government or lobby for forced transaction fees in the leading Bitcoin clients, they will have some power.

Right now, while mining is still profitable at all and we are under a 2 year ROI, there doesn't seem to be any reason not to mine if you can afford to. What ASIC design is going to do is make the design a lot of money. Actually buying the miners and mining once ASIC is out is going to be one of the worst financial decisions you can make (unless you can run an entire farm at the cheapest or free electricity rates).

That must be some sticky stuff you're smoking.

Asics costs a lot to develop, and close to nothing to produce.  Thats a problem, since they are worth a lot at current prices/MH, so there is no reason for a supplier of asics to not price close to market rates. That means there is an enormous gap between initial prices (per MH)  and future prices, because as difficulty goes up, prices per MH will tumble, and difficulty will skyrocket further. An asic vendor has almost unlimited ability to drop prices  to adjust for difficulty,  and it will need to, as difficulty rises. SO he also has an interest in, and ability to push up difficulty by orders or magnitude. That makes it a highly risky purchase for a miner, his investment is bound to lose value far faster than most people will anticipate and probably too fast to make a positive ROI.

Marginal Cost is the change in total cost that arises when the quantity produced changes by one unit (http://en.wikipedia.org/wiki/Marginal_cost -- Wikipedia).  This is the cost AFTER you have paid for your hardware.  For bitcoin, this is the cost of electricity, maintenance (replacement parts), and rent (if you are renting a facility).

The Average Cost includes the cost of the hardware.   For bitcoin, especially with ASIC, it will greatly exceed the marginal cost. Perhaps by as much as a factor of 10-100.

The problem is that the hardware is a fixed cost.  So people will keep making bitcoins so long as the marginal cost is equal or greater to the marginal production.  With market competition, the value of the marginal production will fall until it equals the marginal cost.

I don't know if this sort of thing is possible, but what would be really interesting would be GPU manufacturers entering the game somehow.  They certainly have the means for the R&D whether or not such a thing is a worthwhile project (probably not).  It was interesting and a little bit shocking to see AMD recognize bitcoin mining as a feature of its 7xxx series GPUs - e.g. "Fold and mine faster than ever with AMD App Acceleration powered by the unprecedented 28nm GCN Architecture.".

You have to take the purchase in to account. If it takes 200 years to pay itself back without electricity costs, I wouldnt call that profitable. And if asics get sold anywhere near MH/$ of todays solution, that will very likely be the case.


If you spend $800 million on bailing out banks, thats $40 per coin. So what?

What about alternate blockchains? Can't the hardware be refitted to work on other currencies? Or folding proteins or finding aliens or curing cancer or something?

At whatever price point they decide upon initially, BFL have to sell for a significant amount of money, as they have a substantial NRE to make up. If the price is perceived as too high per MH and GPUs and FPGAs remain too competitive, I doubt they will generate the sales they need fast enough. Remember, BFL have to worry about a potential competitor too, they cant risk spreading their ROI over 5+ years.

So IMHO network hashrate will go up inevitably if BFL is to make a decent return on their investment. Once this effect becomes big enough, BFL will have to lower prices to keep selling boxes, which will further increase difficulty, etc, etc. I dont see how it can be avoided. High initial prices may slow down the effect initially as GPUs are being replaced by ASICs, but if you look in the BFL pre announcement thread, miners seem to be falling over themselves to buy.

More over, high initial prices will also make it even more likely that early adopters will get buried.  If the payback time at current difficulty is on the order of 12 months, there is IMO no chance whatsoever for them to break even - ever. If its much shorter, lots of ignorant miners will buy them in droves even though I suspect 4 months break-even time may still be suicidal. Not enough people truly understand the problem and the difficulty increase will be created by the ones that dont.


I dont think its better for the asic manufacturer to maintain high prices for long. I agree to maximize their profits, they should price them as high as the fool is willing to pay upon release, but once the "biggest fools" stop buying, they will have to drop prices. Not selling boxes is not in BFLs interest either.


I raised the question in the pre announcement thread, and BFL-engineer didnt seem to grasp the problem, or pretended not to. Doesnt bode well. But even if/when BFL sees the problem, its something else to come up with a solution. As long as BFL has a monopoly, they may decide to sacrifice some of their profitability to make sure the market remains healthy: they could impose limits on their sales and not sell more than x TH per y months, even if there is demand. But if/when a competitor emerges, then you simply cant solve it IMO.

I think our only hope is that the ASIC manufacturers think rationally about the future market for their product, and do the following:

Price their respective ASIC's to be cheaper on a $/GH, but not by very much.  This will discourage a mass stampede of miners extrapolating imediate results out to infinity and therefor crushing all mining profitability. 

I think of it this way:  I can buy 100x 5970's and the corresponding hardware for lets say $35,000 and generate 70 GH/s.  Thus I am paying $500/GH.  If I can go buy an ASIC that gives me the same 70 GH/s for $20,000, myself and every other miner of means is going right out and buying it.  In doing so, we ensure our own destruction. 

The only thing that will prevent this is keeping the projected ROI on new ASIC investment relatively close to that of GPU's.  It is better for the ASIC manufacturer (higher profit), as well as all miners (sustainability).  It is a rare win/win.

I hope the BFL boys have thought this through.  Keeping GPU miners around is in their best interests.