BitForce SC - full custom ASIC

[DiabloD3]

Well if it's much less, then break even point is literally 2~3~4 months. Unless these units drastically increase the difficulty, thus increasing break even time.
A Single will then be obsolete and you will have no choice but to trade it in. Rig box will become basic entry point product.

The trade in now looks very generous, even too generous 

My point is that at those prices or even at a much lower price (say $100K) 99% of the buyers of a single will not be able to trade it in, as they won't have the funds or the commitment to put down that kind of extra money. I actually don't see the trade in as a much added value, other than people can perhaps pool their singles, buy 1 ASIC and share the hashing power....

It will be nice for Gigavps, he can probably exchange all his singles, his mini-rigs before they ship, get an ASIC and some money back 

All I can say at this point is that our units will have different models and will be affordable for general public. 100K$ is not affordable
for general public, and is not a good price reference.


Regards,
BF Labs Inc.

I'm assuming you're going to be producing Singles 2.0 and trading those for Single 1.0s.

[drakahn]

Well if it's much less, then break even point is literally 2~3~4 months. Unless these units drastically increase the difficulty, thus increasing break even time.
A Single will then be obsolete and you will have no choice but to trade it in. Rig box will become basic entry point product.

The trade in now looks very generous, even too generous 

My point is that at those prices or even at a much lower price (say $100K) 99% of the buyers of a single will not be able to trade it in, as they won't have the funds or the commitment to put down that kind of extra money. I actually don't see the trade in as a much added value, other than people can perhaps pool their singles, buy 1 ASIC and share the hashing power....

It will be nice for Gigavps, he can probably exchange all his singles, his mini-rigs before they ship, get an ASIC and some money back 

All I can say at this point is that our units will have different models and will be affordable for general public. 100K$ is not affordable
for general public, and is not a good price reference.


Regards,
BF Labs Inc.

I'm assuming you're going to be producing Singles 2.0 and trading those for Single 1.0s.

seems that way, but what are they going to do with the old singles?

[ice_chill]

Well if it's much less, then break even point is literally 2~3~4 months. Unless these units drastically increase the difficulty, thus increasing break even time.
A Single will then be obsolete and you will have no choice but to trade it in. Rig box will become basic entry point product.

The trade in now looks very generous, even too generous 

My point is that at those prices or even at a much lower price (say $100K) 99% of the buyers of a single will not be able to trade it in, as they won't have the funds or the commitment to put down that kind of extra money. I actually don't see the trade in as a much added value, other than people can perhaps pool their singles, buy 1 ASIC and share the hashing power....

It will be nice for Gigavps, he can probably exchange all his singles, his mini-rigs before they ship, get an ASIC and some money back 

All I can say at this point is that our units will have different models and will be affordable for general public. 100K$ is not affordable
for general public, and is not a good price reference.


Regards,
BF Labs Inc.

I'm assuming you're going to be producing Singles 2.0 and trading those for Single 1.0s.

seems that way, but what are they going to do with the old singles?

That's the question I am very interested in 

May sound crazy, but as their ASIC is custom made by them, then possibly they could desolder the FPGA and solder an ASIC in it's place, in which case the Singles could be reused and re-modified.

FPGA can then be resold in the second hand market (as already being done with such chips)

[spiccioli]

Well if it's much less, then break even point is literally 2~3~4 months. Unless these units drastically increase the difficulty, thus increasing break even time.

if half of the available options being discussed in this forum by BFL and others reach the market, difficulty will skyrocket making break even time difficult to achieve and, at the same time, wiping all other kinds of mining rigs.

spiccioli

[Hawkix]

Umm, should be this thread moved to .. well .. new Hardware subforum .. named ASIC?

[DeathAndTaxes]

May sound crazy, but as their ASIC is custom made by them, then possibly they could desolder the FPGA and solder an ASIC in it's place, in which case the Singles could be reused and re-modified.

Unlikely that would assume the custom ASIC is pin for pin compatible with the current FPGA and happens to run at the same voltage.  Even if it did powering (for example) a @10W chip on a 80W DC PSU is going to be horribly inefficient.   Then you consider the heat sink, fan, and other supporting infrastructure is sub optimal it doesn't make much sense.

You take the traded in boards and ship them in bulk to a assembly shop who will remove the FPGA and reball them.  They can also remove any components on the board with significant resell value, scrap the boards, and sell off the components w/ resale value.

FPGA can then be resold in the second hand market (as already being done with such chips)
[/quote]

[MrTeal]

One cool announcement and the mud comes flinging.  I do think it is a bit dangerous to announce the tech so long before it will be ready, but they did pull it off before.  I look forward to hearing more.

Soft launches to make you HARD <grins>

Interesting to see how much of a 'game changer' this is, or if it will just mean an order of magnitude less of consumption for a similarly priced and hashing output.

kind regards

Not much information can be disclosed publicly at this point, although we will do so in the near future.
It reminds me of 1970 where computers could only be shipped by trains and aircrafts. How do they compare
to a single iPad today in terms of processing power, wattage and price? That's where BitForce SC lives...


Regards,
BF Labs Inc.

In 1970 that fastest computer was the Cray 7600 @ a peak of 36MFLOPS, cost $5M and used ridiculous amounts of power. The iPad 3 is capable of 1000x the number of flops, costs 0.01% as much and sips power by comparison. Are you claiming the BitForce SC will bring such improvements?

[DutchBrat]

One cool announcement and the mud comes flinging.  I do think it is a bit dangerous to announce the tech so long before it will be ready, but they did pull it off before.  I look forward to hearing more.

Soft launches to make you HARD <grins>

Interesting to see how much of a 'game changer' this is, or if it will just mean an order of magnitude less of consumption for a similarly priced and hashing output.

kind regards

Not much information can be disclosed publicly at this point, although we will do so in the near future.
It reminds me of 1970 where computers could only be shipped by trains and aircrafts. How do they compare
to a single iPad today in terms of processing power, wattage and price? That's where BitForce SC lives...


Regards,
BF Labs Inc.

In 1970 that fastest computer was the Cray 7600 @ a peak of 36MFLOPS, cost $5M and used ridiculous amounts of power. The iPad 3 is capable of 1000x the number of flops, costs 0.01% as much and sips power by comparison. Are you claiming the BitForce SC will bring such improvements?

Maybe compared to CPU-mining

[ice_chill]

I was thinking about their buy back program:

Assuming the SC is 500Ghs, then a test burn in during one week period would bring in atleast around $10k at current rates (even after electricity is paid for), now because the buyer has already prepaid for the system, they get it for free.

So that money can then be used to subsidize the buy back program, meaning they are only paying a fraction for it.

The real question, is there a time limit to how long after the purchase of you FPGA you can return it for a full refund to fund the SC purchase ?

[BFL-Engineer]

I was thinking about their buy back program:

Assuming the SC is 500Ghs, then a test burn in during one week period would bring in atleast around $10k at current rates (even after electricity is paid for), now because the buyer has already prepaid for the system, they get it for free.

So that money can then be used to subsidize the buy back program, meaning they are only paying a fraction for it.

The real question, is there a time limit to how long after the purchase of you FPGA you can return it for a full refund to fund the SC purchase ?

Regarding Mining, we are not really looking into it. We would've already built our own mining farm if it was
our intention. Our purpose is delivering equipment. The myth about mining to earn money can be said about
any OEM. Our test procedure includes running predefined data and expecting predefined results for extended
period of time, and testing the unit with community-used software just to make sure no incompatibility exists.

The trade-in program involves return of a single, a mini-rig or mini-rig cards purchased from Butterfly Labs.
The date of purchase does not really matter, the credit is given to the customer as soon as the unit
reachs our office.


Regards,
BF Labs Inc.

[P4man]

Assuming the SC is 500Ghs,

Its not going to be 500GH per chip. Although it will depend upon the process they picked and how big they chose to make their chip to optimize yields vs packaging costs, I would guess it will be closer to 10-20GH per chip.
Of course they can put an arbitrary number of chips in a box, making speculating about GH rates useless. What really matters is $/GH. (W/GH will be fantastic anyway).

Now if BFL really manage to be the first to have an ASIC on the market, they can basically price it at whatever they want. Even if they price it similarly to FGPA solutions per GH, people will want it because of the low power. But chip production costs will be completely trivial compared to FPGAs, probably in the single digit $ per GH range; so they just have to price it so they maximize their overall revenue to pay down the NRE.

It will be very interesting to see that unfold; to maximize revenue the sensible thing for them to do is start with very high prices, comparable or even well above FPGA prices, and then just keep dropping them, not unlike for instance Intel would do with their CPUs, launching at $1000 and selling basically the same chip for $15 a few years later.

So Im guessing the price per GH they launch it at, could be cut by one or two orders of magnitude over time.  How fast will depend if any competitors surface. That makes investing in these .. interesting. Assuming they do start with high per GH prices (which I would do), your chances of breaking even on your investment will depend almost entirely on BFLs future pricing strategy. If they drop per GH prices by a factor 10x 6 months later, you are probably shafted. So I wonder if BFL will provide early customers some sort of guarantee?

[BFL-Engineer]

Assuming the SC is 500Ghs,

Its not going to be 500GH per chip. Although it will depend upon the process they picked and how big they chose to make their chip to optimize yields vs packaging costs, I would guess it will be closer to 10-20GH per chip.
Of course they can put an arbitrary number of chips in a box, making speculating about GH rates useless. What really matters is $/GH. (W/GH will be fantastic anyway).

Now if BFL really manage to be the first to have an ASIC on the market, they can basically price it at whatever they want. Even if they price it similarly to FGPA solutions per GH, people will want it because of the low power. But chip production costs will be completely trivial compared to FPGAs, probably in the single digit $ per GH range; so they just have to price it so they maximize their overall revenue to pay down the NRE.

It will be very interesting to see that unfold; to maximize revenue the sensible thing for them to do is start with very high prices, comparable or even well above FPGA prices, and then just keep dropping them, not unlike for instance Intel would do with their CPUs, launching at $1000 and selling basically the same chip for $15 a few years later.

So Im guessing the price per GH they launch it at, could be cut by one or two orders of magnitude over time.  How fast will depend if any competitors surface. That makes investing in these .. interesting. Assuming they do start with high per GH prices (which I would do), your chances of breaking even on your investment will depend almost entirely on BFLs future pricing strategy. If they drop per GH prices by a factor 10x 6 months later, you are probably shafted. So I wonder if BFL will provide early customers some sort of guarantee?

This is not how we have envisioned our production. The strategy you have mentioned is effectively against
the favor of early customers and in favor of late customers. Our strategy has always been in favor of our
early customers.


Regards,
BF Labs Inc.

[Entropy-uc]

http://en.wikipedia.org/wiki/Bait-and-switch

Bait-and-switch is a form of fraud, most commonly used in retail sales but also applicable to other contexts. First, customers are "baited" by advertising for a product or service at a low price; second, the customers discover that the advertised good is not available and are "switched" to a costlier product.

Take out "price" and put in "power consumption". However power consumption is a cost (unless you get free power and then you would use GPUs)

Ngzhang has a lot more trust and respect than you guys will ever have. There is no reason to try to bring down his name. Uncalled for!

In business school we learned the name for people who pay for products that are still in development: unsecured creditors.  Google that to see your future if you are paying BFL 60+ days in advance for products.  This thread is essentially a confession by BFL that they used the cash from mining rig buyers to build ASIC masks.

Honest businessmen go out of business every day for cash flow reasons.  If that business happened to be using your money for working capital, it is gone and you will never see a nickel of it.

This entire market is fascinating to watch develop.  There seems to be a real lack of understanding of basic business and economics in play.  Mining is a closed market.  If I get more, someone else gets less.  And if a lot of other people get more too, I get less than I anticipated and maybe never recover my investment.  

Given stable exchange rates, the total revenue for mining in the next year will be ~$9.5M.  If new investors come in and double the hashing rate they could expect to capture $4.75M of that.  Buying mining rigs at $0.64/GH that would cost $7M.  $7M to capture less than $5M / year, and that revenue stream will rapidly decay as future investments in miners come online.

If ASIC masks really cost $3M it is simply impossible for both BFL and their customers to profit.

[MrTeal]

http://en.wikipedia.org/wiki/Bait-and-switch

Bait-and-switch is a form of fraud, most commonly used in retail sales but also applicable to other contexts. First, customers are "baited" by advertising for a product or service at a low price; second, the customers discover that the advertised good is not available and are "switched" to a costlier product.

Take out "price" and put in "power consumption". However power consumption is a cost (unless you get free power and then you would use GPUs)

Ngzhang has a lot more trust and respect than you guys will ever have. There is no reason to try to bring down his name. Uncalled for!

In business school we learned the name for people who pay for products that are still in development: unsecured creditors.  Google that to see your future if you are paying BFL 60+ days in advance for products.  This thread is essentially a confession by BFL that they used the cash from mining rig buyers to build ASIC masks.

Honest businessmen go out of business every day for cash flow reasons.  If that business happened to be using your money for working capital, it is gone and you will never see a nickel of it.

This entire market is fascinating to watch develop.  There seems to be a real lack of understanding of basic business and economics in play.  Mining is a closed market.  If I get more, someone else gets less.  And if a lot of other people get more too, I get less than I anticipated and maybe never recover my investment.  

Given stable exchange rates, the total revenue for mining in the next year will be ~$9.5M.  If new investors come in and double the hashing rate they could expect to capture $4.75M of that.  Buying mining rigs at $0.64/GH that would cost $7M.  $7M to capture less than $5M / year, and that revenue stream will rapidly decay as future investments in miners come online.

If ASIC masks really cost $3M it is simply impossible for both BFL and their customers to profit.

It's entirely possible that the masks do not cost that much to make. While we'll have to wait and see what happens at the announcement in June, BFL for the longest time swore that they were using custom chips and not FPGAs in their singles. That has been pretty much confirmed to be an outright lie at this point, so their claims of a fully custom ASIC should probably be viewed with a grain of salt.

[bitfury]

It's entirely possible that the masks do not cost that much to make. While we'll have to wait and see what happens at the announcement in June, BFL for the longest time swore that they were using custom chips and not FPGAs in their singles. That has been pretty much confirmed to be an outright lie at this point, so their claims of a fully custom ASIC should probably be viewed with a grain of salt.

Exactly.

My guess is that they have identified an ASIC used in SSL accelerators for e-commerce that is fungible for mining.  Add an FPGA for pre and post processing and you have a mining machine that would be far lower cost than FPGA only. If that is the case, there would be very little opportunity to reduce costs with scale though.  The pricing on the ASICs would be out of their hands and unlikely to decline with volume.

Unlikely. As all of mining stuff should be held in chip. There would be SIGNIFICANT problem of loading/unloading of work at these speeds. Double SHA-256 requires loading of 1024 bits of work and 256 initial state. So say we get 300 Mh/s for Spartan - it will be 1250 bits x 300 Mh/s = 375'000 Mbps = 375 Gbps accelerator chip (if not counting protocol overhead, etc.). Also algorithm in such ASIC would be optimized for SHA256 throughput and not pipelined crunching - that's completely different issue.

About masks costs - that highly depends on tech process they use - as they can go with 180 nm which is cheap, 130 nm which is accessible... 90 nm which is expensive... 45 nm which is out of reach... And with 130 - 90 nm they would be hard to compete with 28 nm FPGA... Because (a) they could keep costs low and produce simple IP with guaranteed result in silicon or (b) they produce best IP for silicon, but risk that they will have several production tries. And overall - you have not mentioned that whole protocol can change, making these ASICs obsolete.

So I suppose first they'll make some money with mini-rigs, then enter NRE costs with ASIC design... and then produce something... nobody knows what it will be however, I think even they do not know :-))))

[Inspector 2211]

Can we order now?

Didn't take you long, did it?

No, seriously, this pre-announcement will pull the rug out from underneath all other vendors - the Icarus/Lancelot guy in mainland China, Stefan in Germany, the British guys, the American guys with their 6500 board, even the genius-level bitstream guys like EldenTyrell and Bitfury.

It's going to be a bloodbath.

[byte1]

I placed an order for a MiniRig just 2 weeks back. Talk about timing 

[smracer]

I wonder if BFL will set a limit on how many AISC's you can buy per month per customer so that no one can just buy up all the production and gain 51% hashing power.

[MrTeal]

It's entirely possible that the masks do not cost that much to make. While we'll have to wait and see what happens at the announcement in June, BFL for the longest time swore that they were using custom chips and not FPGAs in their singles. That has been pretty much confirmed to be an outright lie at this point, so their claims of a fully custom ASIC should probably be viewed with a grain of salt.

Exactly.

My guess is that they have identified an ASIC used in SSL accelerators for e-commerce that is fungible for mining.  Add an FPGA for pre and post processing and you have a mining machine that would be far lower cost than FPGA only. If that is the case, there would be very little opportunity to reduce costs with scale though.  The pricing on the ASICs would be out of their hands and unlikely to decline with volume.

Unlikely. As all of mining stuff should be held in chip. There would be SIGNIFICANT problem of loading/unloading of work at these speeds. Double SHA-256 requires loading of 1024 bits of work and 256 initial state. So say we get 300 Mh/s for Spartan - it will be 1250 bits x 300 Mh/s = 375'000 Mbps = 375 Gbps accelerator chip (if not counting protocol overhead, etc.). Also algorithm in such ASIC would be optimized for SHA256 throughput and not pipelined crunching - that's completely different issue.

About masks costs - that highly depends on tech process they use - as they can go with 180 nm which is cheap, 130 nm which is accessible... 90 nm which is expensive... 45 nm which is out of reach... And with 130 - 90 nm they would be hard to compete with 28 nm FPGA... Because (a) they could keep costs low and produce simple IP with guaranteed result in silicon or (b) they produce best IP for silicon, but risk that they will have several production tries. And overall - you have not mentioned that whole protocol can change, making these ASICs obsolete.

So I suppose first they'll make some money with mini-rigs, then enter NRE costs with ASIC design... and then produce something... nobody knows what it will be however, I think even they do not know :-))))

It's possibly that they wouldn't have a full mask made and have used the funding to produce some prototypes to validate the design and for initial sales. There are a bunch of services that will let you prototype small runs of chips with reasonable setup costs compared to getting a full mask set. One I would have had access to if I was still at University would give me 25 15mm^2 dies on ST's 40nm process for 150k Euros, and the same sized dies on their 28nm process for 225k Euros. I'm sure higher volumes would bring even better prices, but they could have access to the new nodes reasonably cheaply. It's just a matter of when they transition to buying a full mask set on their own.

[BFL-Engineer]

I wonder if BFL will set a limit on how many AISC's you can buy per month per customer so that no one can just buy up all the production and gain 51% hashing power.

We are very aware of the 51% issue and will be
Extremely careful  in this regard.



Regards,
BF Labs inc.