Topic: BitForce SC - full custom ASIC - page 24. (Read 52532 times)

I think BFL has already given us a clue that this will be the case.  They will be selling at least two different products a different price points. If you were to invest a huge sum of money to create an ASIC, you would only want to do it once. Then each product would have a different number of chips on it creating different price points for the consumer. Whether this means 1 or 2 chips or 9 or 10 remains to be seen.

Entertaining how people are pulling random numbers out of a hat for the hashing value of this non-existent chip. BFL, if they are indeed going through with this, could make a large chip or a bunch of tiny chips. All that matters in the end is the $/MH and MH/W ratios for the final consumer product. I personally hope for the AMD GPU route, lots of tiny chips with a scalable architecture. Cluster the chips in manageable quantities with PCB TDPs in a manageable range and scale boxes via daughter cards. This isn't a difficult problem either.. there are no highspeed interconnects needed.

Another lesson taken from AMD's book would be selling chips with DOA cores disabled at a discount. I doubt BFL will have a purchasing agreement to purchase only working dies so, given enough time they'll have a stack of partially working cores. Disable the portions of the chips with the damage and sell an underpowered daughter card as an intro product.

You don't know what the future will hold just the same that I don't know.  But my instinct is that miners won't continue purchasing these things to the point where the original price point will never pay back.  When the payback period is that long, and the margins that slim, people will lose interest on investing their money in mining equipment, reducing the payback period and increasing the margins again.  It's a market that naturally balances itself out.

And how would you know that? The only way to be able to make even an educated guess of your payback time is knowing how many terrahash BFL will sell over that time. Are you saying you have an educated guess for that? Cause I certainly dont.

But I do know the potential is almost limitless, if BFL have a full wafer maskset, they will have a GH printing press with a capacity that far exceeds what you considered possible just a few posts above. You will be utterly at their (and/or their competitors) mercy for you ROI.

True, but that will quickly seem like a drop on a hot plate, like cpu miners exiting 6 months ago.


Im not saying bitcoin wont survive it, Im saying mining profits most likely wont. I already explained my reasoning, Im not going to repeat it.

Time will tell.  I'll just say that I still disagree with you that miners won't be able to pay off their ASICs.

I could see ASIC being disruptive if they get onto the market to early and in big numbers. But unless there are more information about them (Release-date, price , efficeny, production capacity) and even then i wouldn't know how they would/will affect the 'market'.

You forgot to account that the more difficulty increases/less profitable mining becomes the more people will exit it dropping the difficulty and increasing profitability.
And ASIC is not disruptive, it's just the next stage in evolution.

We went from kilohash to megahash to gigahash and survided

So why won't we survive going from gigahash to terahash to petahash ?

Oh yes it can. Easily.
http://bitcoin.sipa.be/speed-lin-ever.png


By my estimate, it takes all of 10 8" wafers on 130nm. That will cost you less than $10K for the chips. You cant even order so few wafers, typically a run will be 25 wafers.

I dont think you fully realize how disruptive asics are.

I think you're worrying a whole lot about nothing.  The potential for future price breaks should already be considered by anyone purchasing new mining equipment.  And even if you're buying equipment that takes a year to pay off, the price/difficulty isn't going to increase 4900% in that year.  Heck, even if the price/difficulty drops in half, you'd still have most of it paid off in one year.

Sorry P4man, I just don't see your doomsday scenario happening.  Most miner's aren't going to waste time with something that'll take > 1 year to pay off, and that'll keep prices and difficulty in check.  And 12 TH/s is a LOT of hashing power to double.

If by pound you mean mm2 die size, you are probably off by more than an order of magnitude. Think 50x faster on a old process and >200x faster at an equal process.

Even if it cost them $0 to develop the asic, the problem remains. Unless you would expect them to give it away for free.


Forget about electricity costs, Im not even factoring that in. That makes the problem for miners worse, not better. I used some numbers to illustrate my point, but the underlying cause remains true regardless of the actual numbers, as long as you assume a trivial marginal cost compared to market value, which is a fair assumption for a bitcoin asic.

The real problem is that BFL or any early asic provider will price their product at a level the market will currently bear. You can argue what that is, but I would say 6-12 months before you earn back your investment seems like a sensible investment, they will sell at that price. But that 12 months ROI, will become 12 years and 120 years as more and more asics flood the market at ever lower prices. How can you know how long it will take before it gets there? As a miner, you cant. You could write a check for 1TH asic miner expecting it to break even in 6 months but never actually get there.


Oh, but Im not worried about BFL. If they are indeed first to market, Im sure they will make a nice pile of money, and they should, they deserve it. Its the miners that will get in to trouble.

Just a guess, but I'd bet that the ASICs will hash at 1GH to 1.2GH each. I know they could be designed larger or smaller, but pound for pound I gathered that ASIC's perform about 3 to 4 times faster than FPGA's. But I'm no expert on the topic.

Put yourself in their shoes. Assume you invested a few million dollar and you now have an asic design that costs $1 per GH to produce. What do you do?

Ill tell you what I would do; I would sell the first systems at a GH/$ ratio thats better than FPGAs, but no more than a factor 2x or so. Why would you?  Better performance per $ and vastly better performance/W, that will make you dominate the market anyway, nothing else would make sense to buy, so you basically capture 100% of the bitcoin mining hardware market. Yet you are selling for say $300 per GH while the chip costs you $1 in marginal cost. But thats okay, you do have to make up for those millions you invested.

Fast forward a few months. You've sold a few terrahash worth of those chips, as a result, difficulty begins to rise, and therefore, demand for your product begins to drop, because the ROI starts taking much longer. What do you do? Simple, you slash prices, cut them in half if you must, its not like your margins wont allow it. And so you sell a few more thousand units at $150 per GH. Until difficulty catches up again and demand slows.... rince, repeat.   until you reach $3 per GH and the bitcoin network is 100x faster than it is now. And early asic customers are making 100x less than they were when they invested.

Now BFL may say they dont want to rip off their customers, and Im inclined to believe them its not what they want to do, but even absent a competitor it will be very hard to avoid doing just that without resorting to a non profit business plan. Factor in a competitor in the above scenario and a price war is a certainty and therefore difficulty will explode even faster and render all those early ASIC miners expensive paperweights that will never earn their investment back.

Unless BFL comes up with some sort of guarantee or promise to limit their sales (!) AND they retain a monopoly,  Id like to short all mining corporations that invest in those early asics. Or later asics for that matter.

Very interesting thread.

I'm watching you closely, BFL 

BF Labs,

Would the performance per dollar of the certain ASIC based models vary slightly, modestly or greatly as compared to other ASIC based models? Granted, these comparative terms are all somewhat subjective. ... For example, the bang/buck of single versus mini is something like ~1.3MH/$ versus ~1.6MH/$. A clear but modest difference - imo.

I suspect the difference will be based on, if anything, a typical quanity discounting curve and not a technological difference. That is, since you probably only have one ASIC chip designed and ready. Is that a correct assumption?

BTW: I appreciate your role in all this. I think the ASIC development is great. It might very well set the stage for the most stable period of mining for all invested miners, I suspect anyway. Like you indicated, nothing will really be the next step in technology anytime soon. Perhaps, the difficulty rating will change more normally - after the ASIC transition/shift.

||bit.out

If anyone receives a broken single, they would definitely send it back to us for replacement. I don't think 600$ of investment
is something people would easily overlook...



Regards,
BF Labs Inc.

Didn't someone send a broken BFL Single to the maker of Lancelot to hack?  ....    ooops