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

[ngzhang]

1, the DSP48As in spartan-6 is very slow.

Oh, one more suggestion that for sure ISE will not synthesize for you.

Adders in Spartan-6 are slow but you have free ROM lying around. How about decomposing the adders? Here's the classroom example:

You need 8+8->8 adder. You decompose it to two halves: (4*4)->(4+1) ROM table lookup and (4+4+1)->4 adder. You could either save on carry-lookahead logic or gain some timing margin that way. This is what I meant by "mangling the design to fit the resources".

This is the type of transformation that I haven't seen anyone trying. I'm not saying that they WILL work in Spartan-6. But they COULD work and they DID work for some ancient FPGA chips.

One thing is for sure: no amount of brute force smartXplor-ing will help to verify that. I'm not blaming anyone for using smartXplorer, but there are just the tricks of the trade that even the oldest computer ponies haven't learned yet.

Edit: Actually the ROM isn't free. You have to carry it around and it costs you by sucking the leakage current. How about putting it to work on the least-significant bits? It's a common sense, really.

not personal assault, but please, please, join this project, write some code, synthesis and run the full fpga design flow, and then review what you type above. maybe you will discover the gap between digital design curriculum and real engineering.

[deepceleron]

I'm glad they announced a public demonstration on the 25th though. At least we'll know approximately by then if it's a scam. Either they demonstrate on or around the 25th, or they are packing their bags right now and will be nowhere to be found on the 25th.

I don't know how a demonstration can be verified. When Steve Jobs unveiled the original Macintosh computer (64k), the demo unit that was up on stage was actually a one-of-a-kind 128k prototype, and the demo used all it's memory. The "$500" demo unit could have $1000 worth of FPGAs on it, or the miner source could have print "hashrate:", hash_rate*2, etc.

[Inaba]

As far as the hashrate being verified, it will be done through my pool.  You might be able to printout a bogus hashrate, but my pool will not report a bogus hashrate as being the same.

For the hardware being something other than what's stated, well that is possible.  

[ngzhang]

I'm glad they announced a public demonstration on the 25th though. At least we'll know approximately by then if it's a scam. Either they demonstrate on or around the 25th, or they are packing their bags right now and will be nowhere to be found on the 25th.

I don't know how a demonstration can be verified. When Steve Jobs unveiled the original Macintosh computer (64k), the demo unit that was up on stage was actually a one-of-a-kind 128k prototype, and the demo used all it's memory. The "$500" demo unit could have $1000 worth of FPGAs on it, or the miner source could have print "hashrate:", hash_rate*2, etc.

you are right. give me 1500$, i can easy build a mining board by using 2X virtex6-240T, which is easily achieve 800-1000MH/s with that PCB size.

[dunand]

The demo on november 25 makes me think about the perpertual motion device that Steorn tried to demonstrate.

[DeathAndTaxes]

you are right. give me 1500$, i can easy build a mining board by using 2X virtex6-240T, which is easily achieve 800-1000MH/s with that PCB size.

Two things.

1) We need to talk.  I would rather have a 800MH to 1000MH board for $1500 than 3 lower priced boards with same output and MH/$ and I am sure I am not the only one.  Would Vertex-5 300 be any good for hashing?

2) Sadly if that is the next step in the scam they likely will get a huge flood of money. "See it works".

[ngzhang]

you are right. give me 1500$, i can easy build a mining board by using 2X virtex6-240T, which is easily achieve 800-1000MH/s with that PCB size.

Two things.

1) We need to talk.  I would rather have a 800MH to 1000MH board for $1500 than 3 lower powered boards for the same price and I am sure I am not the only one.

2) Sadly if that is the next step in the scam they likely will get a huge flood of money. "See it works".

 

1, why a big board is better than some smaller boards?
2, yes.

[fizzisist]

2) Sadly if that is the next step in the scam they likely will get a huge flood of money. "See it works".

Here's the really scary thing: if they're in it for the long con, they might as well sell a few of these boards at a loss to take the money from the hundreds of orders they'll get after a few people report that it works.

Would the power consumption give it away? Inaba, if you do ever get a test unit, make sure to check it with a Kill-a-watt.

1) We need to talk.  I would rather have a 800MH to 1000MH board for $1500 than 3 lower priced boards with same output and MH/$ and I am sure I am not the only one.  Would Vertex-5 300 be any good for hashing?

As I see it, the two problems with this are: 1) economies of scale and 2) if the board fails, you lose a lot more.

[Inaba]

Would the power consumption give it away? Inaba, if you do ever get a test unit, make sure to check it with a Kill-a-watt.

Sure, I will measure power consumption as well.

[P4man]

Here's the really scary thing: if they're in it for the long con, they might as well sell a few of these boards at a loss to take the money from the hundreds of orders they'll get after a few people report that it works.

Why oh why would anyone spend these kinds of resources on some unbelievably elaborate scam that involves selling fully functional $1500 FPGA boards for $599 when they could actually make legitimate profits selling them for $1500?

Guys, seriously. Im all for being careful about possible scams, but is this www.ufos-aliens.co.uk ?

[fizzisist]

Why oh why would anyone spend these kinds of resources on some unbelievably elaborate scam that involves selling fully functional $1500 FPGA boards for $599 when they could actually make legitimate profits selling them for $1500?

I'm not saying they would do it, but it just means that even a few live demos out there still might not be enough to convince me that this isn't a scam.

Along your same line of thinking, and I know this has been asked before, but why aren't they pricing this board at $1500?

[DeathAndTaxes]

Why oh why would anyone spend these kinds of resources on some unbelievably elaborate scam that involves selling fully functional $1500 FPGA boards for $599 when they could actually make legitimate profits selling them for $1500?

Well the build cost would be $1500+.  Sale price would likely need to be $2000+ to cover R&D fraud, retail markup, defects, fixed costs, etc.

At $2K it is comparable to existing units.  Some people like me would be interested simply because I would rather have 10 boards to build a 10GH FPGA cluster than up to 50 smaller/cheaper boards however at $2K it wouldn't be some amazing MUST BUY, GAME CHANGER item.  Sales would likely be very very low.   Most people would keep using existing GPU tech, or smaller cheaper FPGA boards.

I doubt they would sell even 100 in the first year, maybe not even 50.

Another way to look at it ... if you think they are a good product even at $1500?  Why not sell them for $1000 ea?  Hell if they don't sell out they could always drop the cost later.  The $500 pricetag is a call to action.

YOU MUST BUY NOW OR YOUR CAREER AS A MINER IS OVER.  DONT GET LEFT BEHIND .... BUY ... BUY .... BUY.  STOP THINKING.  WE ONLY HAVE 100 YOU KNOW HOW FAST THEY ARE GOING TO GO AT $500.  COME ON MAN $500? YOU CAN'T EVEN GET A GPU RIG FOR THAT PRICE.  HELL SELL YOUR GPUS AND BUY 10 OF THESE.  DONT BE AN IDIOT BUY NOW.   WE ACCEPT IRREVERSIBLE CURRENCY SO BUY BUY BUY TODAY!

[fizzisist]

you are right. give me 1500$, i can easy build a mining board by using 2X virtex6-240T, which is easily achieve 800-1000MH/s with that PCB size.

Wait, how would you put two of these on a board and sell it for $1500?

[DeathAndTaxes]

you are right. give me 1500$, i can easy build a mining board by using 2X virtex6-240T, which is easily achieve 800-1000MH/s with that PCB size.

Wait, how would you put two of these on a board and sell it for $1500?

Digikey prices on those are pure ripoffs.  Note they don't even have them in stock.  They simply buy on-demand and drop ship them to you.  The only reason to use Digikey is because you need 1 for a prototype or R&D and don't want to buy more than 1 unit in case the product fails to live up to expectation.  In even small volume from a real distributor they are more like half that (or less in higher volume).  Digikey is simply taking a 100% markup in exchange for the ability to buy a single unit.

Still I am thinking he was saying build cost is $1500.  Not retail price.  As in the cost of components plus assembly.  Nothing allocated for developers time, development software, retail markup, fraud, defects, capital risk, etc.

[ngzhang]

you are right. give me 1500$, i can easy build a mining board by using 2X virtex6-240T, which is easily achieve 800-1000MH/s with that PCB size.

Wait, how would you put two of these on a board and sell it for $1500?

some days i asked a fpga vender, V6-240T is about 700$

[P4man]

Along your same line of thinking, and I know this has been asked before, but why aren't they pricing this board at $1500?

Likely because they have a significant NRE and a small per chip cost. If they use something like Easypath or hardcopy, their per chip prices could be up to 80% lower than off the shelve FPGAs, but they would have to aim for many thousands of boards to recover the NRA.

[DeathAndTaxes]

1, why a big board is better than some smaller boards?

Scalability.  My "end game" would be putting 4GH/s in a 4U chassis and dropping it in a datacenter.  There it is protected from theft or loss, has reliable electrical power & connectivity.  Also easier to get an insurance policy on a $20K cluster if it is in a protected facility. It would just sit there hashing until the end of time.

Larger cards make that a little easier to achieve than hundreds of small boards.  I agree if the goal is to buy 200MH or maybe even 1GH small cards are "easy", but what about 10GH, 20GH, 50GH?


Would using a backplane work "better"? That would allow better economies of scale and each board would be "cheaper" however you could still get some real densities.

Something like..
http://www.copacobana.org/




Granted it wouldn't be exactly the same.  We are talking about larger FPGAs with higher thermal load so some heatsinks would be required and that means greater distance between cards to allow proper airflow.

Still the general concept would be putting a backplane for multiple cards in a 19" 3U or 4U chassis.  Maybe 18" backplane with 5 cards (10 FPGA) spaced 3" apart.  Then mount two backplanes per 4U chassis (perpendicular).  That would be 20 FPGA.  For LX150 we are talking 160W and 3.5 to 4GH/s.

Not to take this scam thread totally off topic but how much more expensive would it be to adapt existing board so that it can get connectivity (maybe serial/COM ) and power from a backplane.  The backplane then could simply have a PCIex16 connector, a USB controller and communicate to the 10 FPGA via serial bus.  Cooling could be acheived by using 5x 90mm fans at the front and back of the chassis (maybe a third "row") between the two FPGA backplanes.  By moving the USB to backplane that would simplify board design right? Would it be possible to move power supply to the backplane too making each board even simpler.

Am I crazy?  Would it be too much cost/complexity for little gain.  I think it might even be possible to get 3x FPGA per board putting density at 5 to 6GH/s.  If you don't want to derail this thread feel free to drop me a PM.

[O_Shovah]

Just to remind our "modular FPGA miner" project just followed this approach(see the link in my postunderline).
But it seems there wasn't enough interest. We got stuck summer this year.
Im still pursuing it, but alone im not that fast you know

I quoted the copacobana system multiple times.
It was the precessor to the system wich are now produced by SCI www.sciengines.de
I got qoute for that system starting at 12000€.
It is using spartan 6 FPGA's said to deliver 200Mh/s each and 6 FPGa's per board.

so its not that new

[DeathAndTaxes]

Just to remind our "modular FPGA miner" project just followed this approach(see the link in my postunderline).
But it seems there wasn't enough interest. We got stuck summer this year.
Im still pursuing it, but alone im not that fast you know

I quoted the copacobana system multiple times.
It was the precessor to the system wich are now produced by SCI www.sciengines.de
I got qoute for that system starting at 12000€.
It is using spartan 6 FPGA's said to deliver 200Mh/s each and 6 FPGa's per board.

so its not that new

Nice link.  Yeah something like that but with less specialized hardware.  Those boards provide 512MB of RAM per FGPA, DDR expansion, SDHC slot for local storage, high speed busses between FPGA and across backplane, low latency links between servers to build a unified cluster, built in server grade host PC, expansion slots, high speed connectivity to SAN and LAN, etc.

All that adds a lot of cost and is totally useless for Bitcoin.

So take a design like that and make it "ghetto", use a low speed serial to connect the FPGA across the backplane to a USB micro controller, drop everything that isn't necessary and include no host PC.  Just a chassis full of FPGA connected to backplanes.  Then connect power to each backplane and run a single USB cable from each backplane to a host PC (in a seperate chassis).  To expand just keep adding modules until the chassis is full, and then add another chassis.  A single host PC could connect to an entire rack of FPGA chassis.

[P4man]

Scalability.  My "end game" would be putting 4GH/s in a 4U chassis and dropping it in a datacenter.

Am I crazy?  Would it be too much cost/complexity for little gain.

Before you splash that kind of money on expensive and (relatively) power hungry FPGAs, do consider that if bitcoin is here to stay, sooner or later someone will do a (edit) easypath/hardcopy port, or even a  full custom asic. If its not BFL, someone else will, and it will make your off the shelve fpga's look almost as silly as  someone who bought racks full of xeons for bitcoin mining a year ago.

To get a feel for what full custom asics could achieve, have a look here:
http://rijndael.ece.vt.edu/sha3/chip/sha3-asic-datasheet.pdf

If my math isnt off, that test chip gets either 150 or 300 MH/W on an old 130nm process (1.51 Gbps SHA256 for 5mW). On 90nm you might expect double that, on 65nm you might get 4x that. 40 and 28nm, well, do the math, but those are probably too expensive for years to come.