Topic: Is Butter Fly Labs using ASIC's????? - page 2. (Read 5703 times)

Agreed. I doubt there's much gamble. I'm sure they are making a profit. If you forecast Bitcoin, along with mining, along with making money, even the most optimistic would hesitate to predict anything beyond 2 years. Things change so quickly no one really knows what will happen. No one is going to start a business just to get market share when the market might not even exist in a year. I'm pretty confident they're not selling at a loss.

If they put up pre-orders for a new board, they'd get a ton of orders assuming they produce/deliver on the current batch of Singles. I've seen quite a few "I've ordered X amount", factor in people who haven't posted, I'm sure the number of Singles ordered/pending right now is above 100, that's 60K up front. When people will pay and then wait for the product that's a great situation to be in.

I'd guess they can switch gears pretty quick. Assuming they deliver the 100s of orders pending, getting money up front in the future shouldn't be a problem. They managed to double the performance for the same purchase price, I'd lean towards them knowing what they are doing vs getting lucky.

Well seeing as they were able to design a completely new board layout in like month (since power issues cropped up), I would say that that is not likely to be a limitation.

The downside is its a lot of work and somewhat of a gamble but I'm not one to say no to a profit either so I can't blame them for doing what they are doing. Its just one day they are gonna run out of fpgas and then they are gonna have to start over with a new process which may or may not be nearly as cost effective as what they are doing now.

They could sell the boards for more.  I have no idea how much profit they are making but given it is roughly half the $/MH compared to other boards the fact they are selling them for $600 would indicate to me they feel it is "enough".  My guess means that is >$100 per board.  But even if it is only $100 per board there margins would be worse not better using the same chips as everyone else.  Same chips = same performance = only competing on price.  To move 10K units would require cutting margins to nothing.

Would that be better?

As for the economic I have no idea how profitable they are.  Only BFL knows their supply cost and margins.  From a spec standpoint though the board as observed doesn't match an ASIC.  It has a JTAG header and flash loader which only make sense on FPGA.  It runs at 1.1V internal which doesn't match any 45nm FPGA or sASIC.  It gets 10MH/W which is "inferior" to any 45nm FPGA and an ASIC would be many multiple MORE efficient not less efficient.  So no it doesn't make sense that it is an ASIC or sASIC.

I assume they are making enough and if supply is limited they already know this and have planned on moving to another chip when supplies get low.

Ask them...

Some VC's might not need to see a wildly massive profit from a company's trial run before they will commit to lots more cash. (that was speculation btw)

Are they really making that much of a profit from it? They are selling two chips per board, not to mention the rest of the cost of research and development and making the unit. I'd be surprised if they make 100$ profit from each one and their supply of fpgas can't be that great? I mean if they have 2k fpgas thats "only" 100k profit. My point being that's a lot of fpgas to find on the cheap but I guess possible.

Why wouldn't you do it.  They are able to offer almost 2x the performance per $.  I mean when they run out and have say $100K in the bank to spend on 28nm development is that a "bad thing"?  I mean honestly if it was you and you could build "obsolete" boards based on limited supply and easily sell them for a substantial markup and under price every competitor on the market you wouldn't?  Why?  You don't like lots of money?  More orders than you can handle?  Rapid turn around on capital?  Necessary funding to make next gen boards?

I honestly don't get your question.

Would BFL really do that? I mean use old fpgas? Who releases a product with a limited supply of fpgas.

Simple version:
No.  The board layout, power draw, voltage, and presence of JTAG header are all inconsistent with either sASIC or custom ASIC chip.

The chipss are almost certainly last generation (65nm) FPGAs.  Granted a chip w/ 300K LUT tends to go for $1000 or more but if they leveraged prior business dealings they could have gotten a "sweatheart" deal on some end of life chips.  If you look on ebay you can find new but old CPU for sometimes 90% off.  $500 circa 2008 Xeon processors selling for an amazing $48.  The same thing can happen in FPGA world although it is rarely as public.

TL/DR version:
The voltage, power draw, JTAG header, and board layout would be consistent with a large LUT last generation FPGA.

Those aren't ASICs.   An ASIC hashing processor wouldn't be using 2x the wattage (per MHs) as an FPGA.

which means once they run out of this older asic supply the USD price per Mh should move in line with the Spartans.

This explanation is very reasonable to me.

http://www.buyaltera.com/scripts/partsearch.dll

Around $11,000 from the manufacturer

Thats super over priced though I doubt anyone would ever pay that with the new chips out

Two EP3SL340 should do the job. No idea how expensive they are today.

If you would take the time to go through the 2 original BFL threads, you'd realize this has already been dealt with ad nauseum. Of course they are using ASICs. The USB controller is an ASIC. Every FPGA miner is using ASICs if you want to be technically correct. ASIC = Application Specific Integrated Circuit.

As for the 2 large chips with the sanded heat spreader, the best candidate seems to be an Altera Stratix III. Either the EP3SL200 or the EP3SE260. Yes, both of these 'retail' for several thousand dollars. But no one buys these now, and certainly not for retail prices. They are essentially 'end of life' 65nm parts. 2 generations old already. No one uses these for new designs anymore. Why do you think BFLs W/Mhps ratio of twice as bad as anyone using Spartan 6? Because Spartan 6 is a 45nm part, and BFL is using older and power-hungry 65nm parts. But BFL got lucky and found source of cheap, old, last-gen high-end chips. Someone willing to dump them for a low cost. Perhaps from a cancelled project.

Anyway, pursuing this 'question' further is rather pointless. Technically, the answer is 'yes': BFL is using ASICs, but those 2 large chips aren't them. There isn't anything magical about those chips. They are old 65nm tech, but they are high-end. That is why they get the performance they do, but it is also why they use so much more power than any other current FPGA miner on a W/Mhps comparison.

These pics are from the January update
Notice how there doesn't seem to be anything printed on the chips...

Also it uses 2 chips to get a little over 800 mh/s
Thats around 400 mh/s per chip
The spartan 6 chips that all the fpga projects are using get around 200 mh/s per chip and has around 150,000 logic cells

so I would guess you would need around 300,000 to get 400 mh/s per chip

any of the fpga's i've seen with around 300,000 logic cells are in the thousands of dollars range

Also on their FAQ page:
3. Is your system based on FPGA or ASIC technology?

The BitForce processor card is a proprietary implementation of both FPGA and ASIC technology.