Topic: 1GH/s, 20w, $500 — Butterflylabs, is it a scam? - page 3. (Read 123107 times)

The belief in a scam was due to their unbelievable claims.  Claims that turned out to be untrue.

Claimed:
$500
1.05GH/s
19.8W

That's $0.5 per MH and 50MH/W.  Stats that are way beyond what current gen FPGAs can do.

Today we are looking at
$700
~750 to 800MH (simulated performance on static block header)
40 to 50W
That's ~$1 per MH and 15 to 20 MH/W.  Which is inline w/ other FPGA designs and much more believable.

Had they been honest about their performance from the start the reaction likely would be different.  If they can achieve <$1 per MH and ~20MH/W it makes them competitive with other offerings and likely the market leader but it wasn't what was claimed on day 1.

Well they haven't hit 1k mh/s but around 800 if i read right. That's not too bad for a "scammer" with no real product. Keep it coming guys !

a AT45DB642 (U5) tells us a lot. It is a FPGA and needs up to  64Mbits for config data storage. So, it's 1.1V and it's 29*29mm (approx), so, it is altera EP3SL150F780 for near 100% possibility. this fpga's config bit steam is  47Mbit.

or

EP3SL110F780, also 47Mbit


hardcopy is impossible, because they are still working at the code. a hardcopy need a freezed RTL, fully tested.

...s...asic

And perhaps the prototype they gave inaba still had a regular fpga.

how about your cool miner...

My design is DIY; it's $1/MH if you build it at home with hobby techniques and don't
have to pay someone to license the design, populate and solder the board,
test/debug it, etc.

A fully assembled/tested miner w/ 1-2 45nm FPGAs is going to cost more than that.

A fully assembled/tested miner w/ 100 45nm FPGAs can cost less than that.

-rph

  These in reality look to be pulling that 19.8w Per chip, if I am following correctly.

Hmm.  And they got price so low because they acquired a bulk lot of some end of life FPGA.  Plausible but what about power consumption.

It would need to be a pretty large (LUT count) FPGA to push 500MH/s per chip.  No chip that large is going to draw <10W under load.  Not based on 65nm design right?  Of course they would have to know that which makes the power claim silly.  I mean maybe you think "hey we can push this chip to 500MH/s" but then it turns out you can't so you are light on the performance spec.  Not a good idea to promise something you can't deliver but the reason for falling short is certainly plausible right?  However it is harder to explain away a mistake of 100% of the power load.  How does a company get the power draw that wrong?

For the board to pull 19.8W we got to figure PSU inefficiency is going to eat up 15% and regulator maybe another 5%.  That leaves a budget of ~15.8W.  If the fan, and non FPGA components draw 2W our budget is down to ~13.8W.  That's <7W per FPGA.  I don't believe there is any 65nm FPGA capable of the performance they claim that pulls <7W at any price.  Do you know of any?

Based on the silkscreen on one of the test points, they're running at an unusual core voltage of 1.1 volts. There's not a lot of chips that run at that voltage.

What evidence would that be?

28nm FPGAs are apparently starting to become available in small production quantities now for well-connected clients, though good luck getting your hands on any. There's some evidence that the Butterfly Labs boards might be using a 65nm FPGA though.

Statements are submitted by the users and we will do our best to decide with the publicly available knowledge at the event date. I think anyone betting implicitly accepts this. If the company ships to some anonymous people and nobody knows about it, we'll assume it didn't happen. I agree that the specifications are rather strict but it was like that since the beginning and everybody knew that. You are welcome to open a new statement with different conditions if you like.

Did you read the example on our help page about the weighted betting? Simply 45% of the losing bets are distributed to winers proportional to their bets and 45% will be distributed proportional to time weighted bets. 5% goes to submitter, 5% goes to the site.

I don't really presume to speak for BFL in regards to shipping dates or specs, so I can't really comment on that one way or another. As far as I know, they still have a target date for shipping product prior to the end of the year.  

My statement about it being a cluster fuck was more about the conditional issues, such as an independent customer with community trust - who is that?  Do I qualify?  What if i don't decide to purchase a unit until Jan 4th?  What if the first batch goes to private companies/individuals that wish to remain anonymous and we don't see a public unit until later - or alternately, what if they are shipped to a bunch of unknown people with no community trust?  Joe Bitpack can post here saying he got a unit and it works as advertised, but who is he?  Ultimately, what if there is a quantity problem?  If there's only 3 unit available prior to Jan 3rd and all three go to anonymous people - and all three happen to satisfy the requirements - the bet would be false, yet no way to confirm that.  

I'm not saying any of these things are true or that's what will happen, I'm just illustrating the problems with the conditions of the bet.  Also the precision of the bet is somewhat difficult.  What if one measurement is a peak of 1.04 GH/s or max draw of 19.9w?  But if someone else measured it, they'd likely get different numbers.

Conditions should probably be in a range, not specific numbers.  Verification should probably be handled differently (not sure how off the top of my head, though).

That's all I meant.

EDIT -

Also, admittedly, I can't quite wrap my head around how the weighting system works as far as payouts go.

It isn't assumed to be 45nm because of cost.  Eventually 28nm will be the same price as 45nm chips are now and delivery double the performance at half the wattage (roughly).

It is just 28nm FPGA aren't available anywhere for anyone other than test samples to well connected clients.  Even if a company got a test sample they wouldn't get the hundreds required for production.  Even in 2012 supply will be tight for a while.  By well connected clients I don't mean two guys in low-income housing I mean companies w/ decade long history w/ Xilinix and who do 8 figures in purchases each year.

Im sure you could attract the bitcoinica folks - who might actually make  some money on average!

Can you elaborate Inaba? I think the statement is very precise and the company still claims that they will meet the specifications. Did they say anything about the shipping date that I missed?
I know that tension is high due to so much BTC is being at stake but I am doing my best to be fair to both sides.

I see that the demo didn't meet the specifications, but the official statement from BF labs is still that this was a tuned down unit. So at this point I will not close the bets.

Various people suggested switching to intrade type of betting. We believe that it is not suitable for sustaining many small statements. Go ahead and look at intrade. Even they have many empty markets. It is almost impossible to keep every market liquid in that type of betting and illiquid markets are useless. Whereas we can offer statements where only two users bet against each other. Saying that we are considering contract trading where we keep the current betting style and let people trade their bets either if they change your mind or use the advantage of early betting to sell for more.

There are no FPGAs (or much else) available on nodes smaller than 45nm. Its basically only intel and AMD shipping 32nm products, and those are CPUs only.  28nm FPGAs is for next year.

I still wouldnt rule out (65nm ?) s-asic's though.  If it turns out to be off the shelve FPGA's Id really like someone to run some simulations for an altera hardcopy implementation. I know enough people able to invest the kind of money it would require.

When I checked betsofbitco.in a few hours before Inaba's post, the tally was about 275 to 111, so those big bets came in before that. Maybe it was Inaba himself? 

They said before that they will refund bets made after the outcome is determined:


Make that 400 BTC now... crazy!

Question: what are smaller (faster) processes than 45nm? 32, 28, 22?

What is the expected hash rate and power draw from simulations of some of these? It seems that everything is assumed to be 45nm because of cost, but what if...?