1GH/s, 20w, $500 — Butterflylabs, is it a scam?

[DeathAndTaxes]

There are plenty of SHA accelerators, I just havent seen anything in this performance range, or even close to it. That doesnt mean they dont exist, if nothing else, I kind of doubt the NSA uses FPGAs; Id just like to know what shelve to look at

That is kinda the point.
As an example:
SHA-256 is used in IPSEC for example however 1 bitcoin GH =2 billions hashes per second.  @ 512KB per block that is =2tbps (2 terrabits per second).

So nobody making SHA-256 accelerators is going to make one that fast.  An high end IPSEC accelerator might have 1Gbps throughput = 1Bitcoin MH/s.

Same thing w/ accelerators used to offload website login functionality.  Even an accelerator designs to allow 1 million simultaneous logins would only result in 1 Bitcoin KH/s.  Bitcoins usage profile is so off the charts compared to anything else that nobody makes off the shelf chips designs to work on the scale required for Bitcon simply because no market exists for them.

[1713986493]

1713986493

[1713986493]

1713986493

[1713986493]

1713986493

[wormbog]

Happy Thanksgiving, Everyone!

Tomorrow's the big day when Butterfly Labs (or is it BF Labs?) provides a public demonstration of their revolutionary new hashing technology. Either it will turn out to be real, which would be awesome, or it will be delayed or disproven, which will give many of us on this board smug satisfaction. Either way we'll all have something to be thankful for!

In the mean time, enjoy your turkey and stuffing! For those non-American readers out there, get back to work!

[eckmar]

Happy Thanksgiving, Everyone!
...
In the mean time, enjoy your turkey and stuffing! For those non-American readers out there, get back to work!

Hi wormbog,

I am one of these non-American readers, but I will not go back to work. I even wasn't before reading... 
I prefer keep lying in my hammock and wait for the demonstration and proof (or not)...

Have fun...

Sunny regards from the paradise,
 Eckmar

[Inaba]

The software will support Linux, Windows and I believe Mac. 

I am reviewing the driver source code prior to compiling it for tomorrows test as we speak.  I don't know how much or little BFL wants me to talk on the subject at the moment, so I am not providing any additional information.  But to the best of my knowledge at this point, the live demo is still on tomorrow evening CST at my data center.  They've already OK'd me to keep a unit overnight for further testing at another location.  I will inquire as to if they mind if I disassemble the unit and take pictures as well, presuming the tests go as BFL expects tomorrow.

[runeks]

Sweet! Will you be using the unit with a Windows or Linux machine, ie. is it the Windows or Linux driver's source code you've been given?

[worldinacoin]

Is it easy to cascade a few boards together?  Or how many boards can you install on a PC?

[Matthew N. Wright]

Forget these Butterfly lab guys, this one actually resembles a butterfly.

[rjk]

The software will support Linux, Windows and I believe Mac. 

I am reviewing the driver source code prior to compiling it for tomorrows test as we speak.  I don't know how much or little BFL wants me to talk on the subject at the moment, so I am not providing any additional information.  But to the best of my knowledge at this point, the live demo is still on tomorrow evening CST at my data center.  They've already OK'd me to keep a unit overnight for further testing at another location.  I will inquire as to if they mind if I disassemble the unit and take pictures as well, presuming the tests go as BFL expects tomorrow.

I predict some epic crow-eating in 3...2...1...

[uck]

I hope Inaba will make some youtube videos showing the system in operation. That would be great.

[Inaba]

We plan on taking at least some video.  They gave me approval to disassemble it and take pictures (no taking the heatsinks or stuff off, which is understandable - don't want to damage anything.) - so I will be doing that.  It may come down to me doing a physical review of it tomorrow and a live test on Saturday as my schedule is crammed with the holiday, so if the live test slips a day, I would not consider that defaulting on any bets or what have you.

[btc_artist]

The software will support Linux, Windows and I believe Mac.  

I am reviewing the driver source code prior to compiling it for tomorrows test as we speak.  I don't know how much or little BFL wants me to talk on the subject at the moment, so I am not providing any additional information.  But to the best of my knowledge at this point, the live demo is still on tomorrow evening CST at my data center.  They've already OK'd me to keep a unit overnight for further testing at another location.  I will inquire as to if they mind if I disassemble the unit and take pictures as well, presuming the tests go as BFL expects tomorrow.

Have you signed an NDA with them?  Have you met them in person yet?

[btc_artist]

Not implying this is the case AT ALL, but wouldn't it be an epic scam if Inaba gave a down-to-earth review proving that the magic box works as described, several hundred orders roll in at full price, then it turns out he was in cahoots.....  Again, I'm not even suggesting this is the case, I just let my mind wander.

[gmaxwell]

Not implying this is the case AT ALL,

Not implying this is the case AT ALL, but wouldn't it be epic if it turned out that the mining circuit was acausal logic implemented by arranging the gates in a closed time like curve that allow the system to only attempt solutions which are valid and thus actually achieving infinite speed. ... but then Butterfly labs couldn't productize that because it would make bitcoin worthless, so instead they rate-limit it to realistic speeds in order to maximize their sales. Again, I'm not even suggesting this is the case, I just let my mind wander.

[btc_artist]

Not implying this is the case AT ALL,

Not implying this is the case AT ALL, but wouldn't it be epic if it turned out that the mining circuit was acausal logic implemented by arranging the gates in a closed time like curve that allow the system to only attempt solutions which are valid and thus actually achieving infinite speed. ... but then Butterfly labs couldn't productize that because it would make bitcoin worthless, so instead they rate-limit it to realistic speeds in order to maximize their sales. Again, I'm not even suggesting this is the case, I just let my mind wander.

Nice.

[P4man]

I predict some epic crow-eating in 3...2...1...

This. About time too. I hope the debate will now shift to the more interesting questions:
- how are they achieving this performance? Structured Asic a la Fast Path is my guess but I wonder if anyone sees any alternatives. Maybe we should open a betting pool for this :p.
- how long until GPU mining is dead? Even for those with "free" electricity it almost makes more sense to buy this BFL product instead of GPUs. Assuming BFL is doing a first production batch of 1000 units that they can assemble and ship over the next few months,  thats already 10% of the network. And thats ignoring the "rig box". Just 100 of those would equal nearly 50% of the bitcoin network. I guess I have a few months to get rid of my GPUs  

[P4man]

Do the math.  Even with their specs, at the stated price it takes about a year to break even at current difficulty and BTC price. Thats a huge cashflow issue for a startup and risking the survival of your company on mining profitability and BTC price is a gigantic risk. Then there is the risk of someone else coming out with a better product, like custom asic, within the year, rendering your expensive mining farm almost worthless.

Compared to being able to sell those devices with perhaps 50% gross margin (just a guess if they are using s-asic) its a no brainer.

[P4man]

Keep in mind their costs will be less that 50% gross margin.    

Im comparing revenue. Even if they cost BFL $0 to produce, it still takes a year at current price/diff to earn them as much mining as selling a unit would earn them.

I could give one to a guy I know and have him mass produce it (odds are even cheaper) in China and then have them send it back to me.

Interesting. Please send that guy a Core i7 and let him mass produce it for me

Hint: the cost is not in the PCB or assembling. Thats maybe $10-$20 each. Assuming those are s-asics, the cost is the NRE they have to pay Altera for the masksets, IP etc. "A guy" in China cant do that. If they are regular FPGAs, or some other off the shelve chip, obviously the cost is in acquiring those.

[P4man]

I am assuming that BFL does not have anything too special due to the price they are selling it at and the fact that most of the people interested in this are bitcoin miners.

On the contrary, the price (and performance) strongly suggests they do have something "special". Either they had some brilliant idea to make off the shelve FPGAs vastly more efficient, in which case, someone else is going to be able to do that too,  or they have a custom chip. Custom chip means very low per unit cost compared to FPGAs, but a fairly high fixed cost. That means you want to sell truckloads. In either scenario you want to sell as many as can as soon as you can, and not wait 12 months hoping to still be king of the hill so you can just break even.

If there was a side market for this like gaming for the GPUs then sure.

Their website suggests there are other applications for their product. Code breaking would be one obvious candidate, but they mention medical imaging, finger print processing and I read somewhere monte carlo simulations too. If true, that suggests their chip is more general purpose than SHA hashing alone, but we will shall see how that pans out.

At current mining conditions it would take them about 7 months to break even assuming their cost is $450. After that its pure profit. Yes, Conditions will change but...

Its actually 9 months at that cost, only to break even on per unit cost, ignoring R&D, overhead etc costs, the need for a large and secure datacenter, electricity etc etc.

But thats not even the real comparison, its how long it takes (and at what risk) to be more profitable than selling. Even assuming your BOM estimate, and the $599 list price, its still > 12 months before they would make more money mining than by selling the hardware. And at a huge risk. It could easily be 24 months or never.  No sane investor would accept that risk when there is such an easy and almost risk free way to get a nice positive ROI way sooner.

Moreover, there is no reason to assume development will stop after the single board and that will be the best and only ever product in the market for 12 months. Even if BFL themselves have no plans for something better in 12 months, there are other companies claiming to work on full custom asics to be launched in 2012. If thats true, BFL had better made back its investment with a nice profit before that hits the market, as it would crush the single board and make it unsellable for btc mining.

This is the risk you want to offload to miners that invest in these devices. Its the reason I am not going to buy a BFL single board even if they achieve all their promises. IN 6 months its entirely possible it will prove a cost I can never recover. And at best in 12 months Im only just breaking even.  If Im not willing to risk $600 on that, its hard to see how a company and its investors would be willing to gamble their existence on it.

[fred0]

Also just because their website says there is other demand does not mean it is true. Like I said, I highly doubt the majority of these things will be sold to anyone other than bitcoin miners.

Honestly my 1% thinking that this might not be a scam is that they thought of something clever and that it can be copied. I doubt they ordered 10,000 chips yet have so many other failings business wise.

I seriously doubt that the rig box or "supercomputer" on the BFL web site is primarily for bitcoin.  Just my hunch, but these seem geared to military/intel customers.

[fred0]

I know that is what the website would lead you to think however in my experience military contracting does not work this way.

I honestly do not buy it. Just a hunch.

Provided proof that this tech works, how may rig boxes are you going to buy?

Pre-orders start TONIGHT!