In the long run yes however I was more looking for the short term. Once shipping in volume margins on hardware is going to collapse and trend towards a small markup over cost. Still even if raw silicon is $0.20 per GH/s the overall system is going to be higher. Even simple stuff like high quality DC to DC PSU aren't cheap and can easily be more than the ASIC itself. Throw in AC power supply, case, cooling, testing, assembly, labor, etc and you will be hard pressed to build a system below $1 per GH/s.
Topic: Guesstimate thread for total ASIC pre-order hashing power. - page 3. (Read 7919 times)
Vendors who are past their NRE cost can provide practically unlimited quantities of chips at the cost that is orders of magnitude below today's prices. Thus, the answer to your inquiry is mostly in the domain of psychology of the miners, not in the domain of engineering of the vendors.
The hashing power that will be coming online through the gaping anus of all these preorders is elastic in the sense that vendors will simply keep selling as long as (pre)orders are coming in. Once the supply of suckers starts diminishing, the vendors will simply drop the $s/hash ratio. This will attract new customers, and also force some of the old ones to keep ordering so they don't drop out of the game (monarch anyone?). Then the price drops again. And again. And again. Only when we get close to the production cost, and vendors start operating on thin margins, will the game of power efficiency begin.
What does it cost to produce high volumes of ASICs? A couple of tens of dollars per chip, which is thermally limited to a couple of hundreds of watts, which currently might provide a couple of hundreds of GHash/s. There you go. Let's say ~0.2$s/Ghash. Next, we account for the RoI period miners are comfortable with - a year maybe? That's 1.3M coins up for grabs. Their value is anyone's guess, but let's say $400 per coin in the next year or two. That's half a billion dollars to be made, ignoring fees. The miner's investment then translates into 2E9 Ghash/s. Give or take. Once we hit that, the J/hash will become more important, and that's a whole another amazing story full of exotic technologies, exotic places, and politics.
The hashing power that will be coming online through the gaping anus of all these preorders is elastic in the sense that vendors will simply keep selling as long as (pre)orders are coming in. Once the supply of suckers starts diminishing, the vendors will simply drop the $s/hash ratio. This will attract new customers, and also force some of the old ones to keep ordering so they don't drop out of the game (monarch anyone?). Then the price drops again. And again. And again. Only when we get close to the production cost, and vendors start operating on thin margins, will the game of power efficiency begin.
What does it cost to produce high volumes of ASICs? A couple of tens of dollars per chip, which is thermally limited to a couple of hundreds of watts, which currently might provide a couple of hundreds of GHash/s. There you go. Let's say ~0.2$s/Ghash. Next, we account for the RoI period miners are comfortable with - a year maybe? That's 1.3M coins up for grabs. Their value is anyone's guess, but let's say $400 per coin in the next year or two. That's half a billion dollars to be made, ignoring fees. The miner's investment then translates into 2E9 Ghash/s. Give or take. Once we hit that, the J/hash will become more important, and that's a whole another amazing story full of exotic technologies, exotic places, and politics.
2. MetaBank will receive 1400 boards in 8 days.
—> 1400 * 8 * 2.9Gh/s = 32.48Th/s
Meaning, don't forget the Russian BitFury projects.
Added. Thanks.
1) Has anyone compiled a guestimate of the amount of pre-orders.
2) I did a breakdown in a previous thread I cannot find ATM. Worked out to an average .1TH per BFL order and ~70k orders using the ~4900 known orders.
3) Refunds were not taken into account due to lack of information.
4) Monarchs were considered the "and rising"
2) I did a breakdown in a previous thread I cannot find ATM. Worked out to an average .1TH per BFL order and ~70k orders using the ~4900 known orders.
3) Refunds were not taken into account due to lack of information.
4) Monarchs were considered the "and rising"
For posterity.
1) Has anyone compiled a guestimate of the amount of pre-orders.
2) I did a breakdown in a previous thread I cannot find ATM. Worked out to an average .1TH per BFL order and ~70k orders using the ~4900 known orders.
3) Refunds were not taken into account due to lack of information.
4) Monarchs were considered the "and rising"
2) I did a breakdown in a previous thread I cannot find ATM. Worked out to an average .1TH per BFL order and ~70k orders using the ~4900 known orders.
3) Refunds were not taken into account due to lack of information.
4) Monarchs were considered the "and rising"
For posterity.
or if 70% of the estimate gets delivered we will be around 400M diff near the first of the year?
My projections have us hitting an average diff of 411,484,534 in December 2013. This would be an average network speed of 2.94Ph/s and would mean that most ASIC equipment out would be unprofitable to mine with shortly afterwards (2 to 3 months) unless there is a price spike or electricity and hosting get a great deal cheaper.
I can't find a source to back up 1000th for asicminer by the end of the year, I am sure I read it somewhere, but it must not have been official. A lot of people will have alotmore to say than me, but a few points:
Friedcat seems pretty motivated to keep a roughly 20% share of the hashrate. While they might have dropped to 10% lately, I am pretty confident they are ready for what's to come.
Friedcat would have access to more money than any of the other manufacturers.
The next generation of their hardware is known to be in the works.
They plan on leasing out their hardware, which I would assume means they will have a lot of it.
And one last bit of evidence would be their share price. If everyone lost confidence in friedcat's ability to deliver, their shares would be worthless. They have dropped recently but only because of their recent lull.
Friedcat seems pretty motivated to keep a roughly 20% share of the hashrate. While they might have dropped to 10% lately, I am pretty confident they are ready for what's to come.
Friedcat would have access to more money than any of the other manufacturers.
The next generation of their hardware is known to be in the works.
They plan on leasing out their hardware, which I would assume means they will have a lot of it.
And one last bit of evidence would be their share price. If everyone lost confidence in friedcat's ability to deliver, their shares would be worthless. They have dropped recently but only because of their recent lull.
2. MetaBank will receive 1400 boards in 8 days.
—> 1400 * 8 * 2.9Gh/s = 32.48Th/s
Meaning, don't forget the Russian BitFury projects.
1) Has anyone compiled a guestimate of the amount of pre-orders.
2) I did a breakdown in a previous thread I cannot find ATM. Worked out to an average .1TH per BFL order and ~70k orders using the ~4900 known orders.
3) Refunds were not taken into account due to lack of information.
4) Monarchs were considered the "and rising"
2) I did a breakdown in a previous thread I cannot find ATM. Worked out to an average .1TH per BFL order and ~70k orders using the ~4900 known orders.
3) Refunds were not taken into account due to lack of information.
4) Monarchs were considered the "and rising"
Nice analysis. Much better than the "let's add 50% every month until forever".
Curiously, if you want to choose a single value to estimate beginning of January hash power, 50% per month is not nearly enough
Assuming 50% till January:
500*1.50^4. = 2531.25
To get a value close to the sum (assuming "only" 500 TH/s for KNC) to be delivered till January you need about 68%:
500*1.68^4 = 3982.97
But choosing a single value is not the purpose of the topic, sorry.
Nice analysis. Much better than the "let's add 50% every month until forever".
Was talking just about the KNC numbers.
How do you get 2000 as the limit, when the running total is 3900, or did i miss something
Wow! Sweet analysis DeathAndTaxes. Those extra considerations do constrain the die size quite a bit more to the upside. 2 PH/s is definitely an upper limit likely to be much above the shipped total (nevermind 3). Learning quite a bit here. 
DT, thank you for this post... your one of the few 'givers' here.
at 3900 Th, that puts diff at 540M
So, can we say that if everyone delivers, we will be at 540M near jan 30 2014?
or if 70% of the estimate gets delivered we will be around 400M diff near the first of the year?
edit: I personally have been doing my guesstimates for profit starting Jan 1 at 2500 Th, adding about 300Th week. Looks like i will have to bump my starting point on jan 1 to 3000Th
at 3900 Th, that puts diff at 540M
So, can we say that if everyone delivers, we will be at 540M near jan 30 2014?
or if 70% of the estimate gets delivered we will be around 400M diff near the first of the year?
edit: I personally have been doing my guesstimates for profit starting Jan 1 at 2500 Th, adding about 300Th week. Looks like i will have to bump my starting point on jan 1 to 3000Th
Did the bitfury numbers include the 100TH (now 200TH) mine project?
No. Will add.
If their die size is indeed that small then 2 PH/s seems likely but I don't think the die will turn out to be that small. I will make a note in the OP and let others weigh in though. Given the use of 55mmx55mm package and the power consumption I believe the die is much larger at least 20mm x 20mm and that is being conservative. That would put it at more like 15 TH/s per wafer or ~375 TH/s per batch. Two runs would be up to 750 TH/s although my guess is the die is larger than 20mm x 20mm.
The package:
They are using a 55mm by 55mm package. With flip chip packaging it is technically to fit an 11mm die to a 55mm package but would be rather bizarre. Yes using a larger package helps heat dispersion but it has diminishing returns otherwise CPU and GPU would have giant packages the same size as the heatsinks they ultimately will need.
The power:
The quote estimates size by using BFL logic and shrinking one core from 65nm to 28nm and then scaling out. That assumes KNC will have comparable GH/mm2 as a 28nm BFL but if true they should also have comparable GH/W. However BFL is 6W/GH (at 65nm) and scaled down would probably <1W/GH, yet KNC is reporting 2.5 GW/GH. There isn't a 1:1 relationship but power consumption generally scales with die size. Another data point would be HF which has a 18mmx18mm die (28nm) which is ~1/3rd the quoted size yet has a power consumption of 0.6 GH/W. To believe their chip will be 1/3rd that of HF but use 4x the power seems less improbable.
The package:
They are using a 55mm by 55mm package. With flip chip packaging it is technically to fit an 11mm die to a 55mm package but would be rather bizarre. Yes using a larger package helps heat dispersion but it has diminishing returns otherwise CPU and GPU would have giant packages the same size as the heatsinks they ultimately will need.
The power:
The quote estimates size by using BFL logic and shrinking one core from 65nm to 28nm and then scaling out. That assumes KNC will have comparable GH/mm2 as a 28nm BFL but if true they should also have comparable GH/W. However BFL is 6W/GH (at 65nm) and scaled down would probably <1W/GH, yet KNC is reporting 2.5 GW/GH. There isn't a 1:1 relationship but power consumption generally scales with die size. Another data point would be HF which has a 18mmx18mm die (28nm) which is ~1/3rd the quoted size yet has a power consumption of 0.6 GH/W. To believe their chip will be 1/3rd that of HF but use 4x the power seems less improbable.
KnC ASIC will have about 100 GH/s realised with 192 engine IPs (see KnC R&D news from 7/19/2013).
I assume that an engine IP is equivalent to a pipelined hash core. To realise 100 GH/s they must run at a little bit more than 500 MHz, what is feasible in 28nm.
Based on technology scaling (2x more logic on same area form technology node to next technology node) we get based on BFL hash core size (estimated based on BFL die including 16 cores removing spare area and pad frame overhead) from 65(55)nm -> 45(40)nm -> 32(28)nm an estimated area per hash core of 0.6 mm2 in 28nm.
This results in an overall KnC die area of about 115 mm2 (0.6mm2 x 192).
Adding some area for supporting logic, I would say max 120 mm2 would be a good target.
An 28nm 300mm wafer has an area of about 70000 mm2. So we get 583 dies per wafer (assuming 100% yield, not realistic I know, but you can scale it yourself to your yield assumptions).
For 50 wafer:
583 dies/wafer -> 58.3 TH/wafer
50 wafer -> 2915 TH
Minimum ordering at foundries is normally one lot (25 wafer). Maybe also half lots are possible. But keep in mind, additional wafers costs nothing compared to the initial mask costs.
So would you say the above is too crazy?
Did this wafer reasoning apply to other manufacturers?
Maybe they ordered only one batch of 25 wafers which would be only 1.5 PH/s and thus only 3 times your estimate. Lets say there's a lot of inefficiencies and put it at 1PH/s? 50 wafers plus inefficiencies (66%) ~ 2 PH/s ? Been looking forward for a while to get someone to comment on the wafer estimate by HyperMega.
Did the bitfury numbers include the 100TH (now 200TH) mine project?
Didn't hashfast partner with icedrill for 250-363th for November? (https://bitcointalksearch.org/topic/bitfunder-icedrillasic-ipo-235-thash-mining-operation-powered-by-hashfast-269216)
Friedcat, he has more money than anyone, has said he would have a petahash ready for the end of the year. (Looking for source).
Friedcat, he has more money than anyone, has said he would have a petahash ready for the end of the year. (Looking for source).
Good point I hadn't really considered any other mining bonds/companies because they are generally already included in miner/chip sales however this is an exemption. I added the lower end of their projected range to the OP.
Jump to: