You are correct, but that effect is so marginal, particularly for bitcoin asics running at 100% load, so dynamic power is so much more important than leakage, it can be safely ignored in this context.

~7000 bitcoins gone.

ghash is no longer reporting their hashrate. Maybe its a bug because they passed 1PH, but I also noticed the past days there where wild swings, hashrate would drop 500TH or more. Could be technical issues, but after reading this, I do wonder if its not something more nefarious.

Anyway, regardless if there is something going on, Id like to urge miners to point their hashrate to some smaller pools than ghash and btcguild. Its just not good for bitcoin to have 2 entities with almost 30% of the hashrate each, and this gives more credence to the "selfish miner attack" recently published.

BTW..


One has to wonder how they ever expected to ship miners a few weeks before receiving wafers. Didnt they always claim the silion production was on schedule?

You are misreading. Submersion cooling doesnt magically make the chips more efficient, it makes the cooling potentially much more efficient. If you dont need AC, cooling is just the fans and maybe 2% of the total power consumption, reducing that by "97%" means you drop overall power consumption by, well almost 2%.

If you compare it to AC, it all depends on what AC system you compare to. Savings there can be substantial, easily 50% if you assume inefficient domestic AC systems (a lot less for typical heatpump cooling in a datacenter), but thats still nowhere near enough to make a chip that uses 1000% more W per GH suddenly competitive.

Not too mention the fact anyone can buy submersion cooling, its not like AM invented and patented it. For instance:
http://www.grcooling.com/


Efficient? Now you confuse the market price with efficiency and cost. AM cant charge more per GH than KnC. Thats not good for AM, thats BAD.  Because the fact is their current chip produces 0.3GH where the competition provides 100+GH and soon 400+GH per chip. You really think AM can produce their chips 1200x times cheaper than their competitors? Of course not, so soon they can no longer sell anything based on that chip at a positive margin.

How does efficient cooling help reduce chip prices? To match one KnC asic, AM needs >1500 gen 1 asics and almsot 10x as much electricity. Even if the cooling is 100% free and 100% efficient,  there is no way they can become competitive without new chip.

That said, it does look cool. Im just not yet convinced it makes financial sense, nor do I quite see how AM is going to monetize this.

Im surprised this is not causing any more discussion. The largest pool appears to be double spending, and everyone is mum? Has this been debunked or what?

keep in mind a bitcoin asic almost certainly requires less metal layers than your typical soc/cpu/gpu, which will decrease fabrication time proportionally. 24 days sounds very fast to me too, but I doubt that number came out of nowhere. I actually suspect its the fabrication time, not including the tape out process and mask generation, but FWIW.

According to BFL, a "bullet run" at globalfoundries takes 24 days. To quote Josh "The initial process is what is called a bullet run, it will allow us to get a finished set of wafers in approximately 24 days from the date we tape out."  At TSMC afaik hot lots take longer than that. Does anyone know where CT is fabbing?

As for what BFL actually achieved, if you are referring to their 65nm chip, Im quite sure they failed their first tape out attempt (the "refraction issue"), and had to redo their physical design flow. I dont think they told us when they successfully taped out.

Going out on a limb, Im going to hypothesize the "statement" they are preparing is nothing else than a working prototype. And thats why they remained silent and are running late, they are hard at work and want to show the thing working.

Am I betting money on this? Nope, but its just as plausible as all the other theories.

Here is another source:

From the economic perspective, data from International Business Strategies shows that the move to 20nm and FinFET results in essentially the same cost per die (Q1 2014 estimates), especially as devices increase in size (see fig 3).

- See more at: http://www.newelectronics.co.uk/electronics-technology/what-makes-finfets-so-compelling/56795/#sthash.2brGAsMY.dpuf

The text not withstanding, For small dies, such as those used for bitcoin, the per die price difference is still substantial. However, do note the absolute price levels of just ~$7-8 for a 100mm² 28nm chip.  

There is the possibility of higher power efficiency, which could allow you one day to sell those 20 or 16nm chips at >$2.5X whereas you might sell close to nothing at $X.
Now I also doubt the volume would be there to warrant the investment, even assuming these shrinks are significantly more efficient. It probably makes more sense to undervolt 28nm chips and package more of them.

Then again, if by promising superior efficiency and performance,  you can presell enough of them to cover the NRE; why not do it? You'd have a maskset that will last you "forever", and if you are the first one to gather enough presales, you may end up being the only one to do it.

BFL's schedule is more or less realistic (if executed by someone else ).
Wafer production can be accelerated if you pay for it. BFL themselves claim to do that for monarch, reducing the time to 25 days (according to them).
Cointerra is also rumoured to have paid for this for their first batch, no idea about HF, but its irrelevant since their wafers are already done.

2 weeks for packaging seems very long to me, unless they send the wafers by slow boat from Germany to Asia and get the packaged chips back to the US. The actual packaging process itself is almost instantaneous.

As for the FUD about their experience. I really have to laugh at that.
Here is my interpretation: "We fucked up the first tape out attempt because we didnt have a clue and we had never done full custom asic design. We learned a few new words in the process, like "refraction". Then we fucked up our packaging choice because we fucked up our power consumption estimates and we just ordered the cheapest possible packaging. Even when we selected the correct package type appropriate for the power consumption, we still managed to fuck up the underfill. And once we finally got the chip to work, we had all sorts of issues with the PCB, cooling, firmware. It took a year longer than expected, but now we are the experts because we made every mistake there is to make in this industry".

Makes me chuckle. As KnC and many others proved, if you know what you are doing, none of those are hurdles one expects to bump in to. Bumping in to all of them just proves BFL hired completely incompetent engineers or design house.


Thats always possible, but unlikely, and nothing suggests this is this case, and unless they meanwhile solved the substrate sourcing issue, they wouldnt even know yet.
FWIW, even BFL didnt need a respin to get their asic working.

But you are not measuring what you claim that you are measuring. The "cost of mining bitcoins"  != hardware prices per GH, and especially not ones sold to end users.
Those are market prices formed by supply and demand, and demand is correlated with bitcoins price and expected future difficulty. Is that really so hard to grasp?

Here is a hint for you: those hardware vendors are making a fortune. You speak of "cost", what do you think it costs KnC to produce a GH? What would it cost Bitfury? Or why is that less relevant than what it costs you?

Secondly you can measure any correlation you want, but you are drawing the wrong conclusions from it (and from incorrect data) because you are (deliberately?) oblivious to the underlying mechanisms, which really arent very difficult to understand.

Does anyone have an educated guesstimate of what the NRE for a 20/16nm asic would be?

You are not tracking the cost of mining 1 BTC. You are looking at the purchase price per GH. How many bitcoins that purchase will actually mine is utterly dependent on the network growth, which is (insufficiently) priced in the selling price, and which you are completely ignoring.  So you are ignoring by far the most important mining variable today, and pretend it will be no different in 12 months when hashrate stagnates,  as today when mining revenue is more than halved every month.

You are also ignoring what will become the single most important mining variable in a few months: electricity cost. Once network growth begins to stagnate (or keeps pace only with BTC pricev) you can write off your hardware purchase over 10 years if you want to, completely throwing your only metric out of the window,  but you still have to pay the electricity bill every month. And it wont be long before the cost of mining (actual cost, ie, electricity cost + hardware write off)  will roughly equal the mining revenue. As long as its more than marginally profitable to mine, more people will want to do it, and thus difficulty will go up. Its really that simple.

On top of that, a few things are special about todays situation:
- exponential network growth making estimating mining revenue almost impossible
- due to that growth, extremely short hardware write off period
- inability of suppliers to meet market demand, ie, ridiculous high hardware prices and almost irrelevant electricity costs.

None of these factors will remain.

Whatever it is you think you are measuring or trying to predict, without taking any of the above in to account, it has zero meaning.

So what is it, they dont and wont have chips, or they are selfmining? It cant be both.

But for how long? As the network hashrate grows exponentially, the value of these asics drops proportionally. Even the current rise of BTC value is almost meaningless compared to that.  IMO two things can happen; either these companies will start accepting preorders for 20nm chips in the next 3-4 months, or these chips wont materialize.
If they go down the early preorder route, they can lock in the NRE at no risk (other than to the miners who will be screwed again). If they dont, 3-4 months from here, my guess is mining profitability and asic margins will have dropped so low, its probably not worth it anymore.

Actually its the opposite. Increase of bitcoin exchange rate means that more even hashing power will be sold and you will mine less BTC, and mining BTC is all the machine does and what you bought the machine for. If you wanted a dollar profit speculating on a BTC price increase, you should have bought (or held)  BTC.

$4M is about what it takes to get a 28nm design and maskset. Of course the problem is that if they have not spent it yet, they will end up last (or second last, there is always the possibility of beating BFL ) behind KnC, HF, CT, ActM, Bitmine, BlackArrow, .. possible BFL, Avalon, Bitfury,. By the time it could hit the market, market will be so cramped and margins will have eroded to the point where it might become difficult to just recover the NRE.


I cant be bothered to check the math on that, but even if true, who cares? It just shows what the market price per GH is roughly. The real problem is of course that AM needs like 1500 asics to match one Jupiter.  Unless you think Friedcat can produce these chips for $0.1, thats not good news.