1GH/s, 20w, $700 (was $500) — Butterflylabs, is it for real? (Part 2)

[BFL]

Don't suppose anyone would be willing to give me a copy of the protocol specs and access to a box or VM with a board connected? I already have a poclbm-based FPGA mining client lying around that could almost certainly be adapted. (It does fancy things like pool failover, roll-ntime support and testing found nonces against the previous work unit as well as the current one for increased efficiency.)

Please email me at office @ butterflylabs and I'll set you up with the protocol and a machine hosting one of the test units.  FYI - we will be publishing the protocol soon as well.

Regards,
BFL

[1714077327]

1714077327

[1714077327]

1714077327

[1714077327]

1714077327

[1714077327]

1714077327

[BFL]

"end of our development process"

I hope you meant you will redesign this, get the power fixed. Get the temps down...

Yes, in case this was not clear...   this modification has already been made and production units are currently underway with twice the power capacity.

[BFL]

You already redesigned it? What modification? Just the power box change? I'm lost...

The external PSU is not a manufacturing issue.  The power system of the PCB has been doubled in capacity to ease the bottleneck caused by the power needed by the processor chips.  This is what we've been working on over the last two weeks.  

[rjk]

You already redesigned it? What modification? Just the power box change? I'm lost...

The external PSU is not a manufacturing issue.  The power system of the PCB has been doubled in capacity to ease the bottleneck caused by the need for power from the chips.  This is what we've been working on over the last two weeks. 

From what I understand, some of the additional power consumption is because of the power circuit being run above design levels. Once the power circuitry is oversized, the efficiency should increase, and the total consumption should decrease overall. Is this correct?

[BFL]

You already redesigned it? What modification? Just the power box change? I'm lost...

The external PSU is not a manufacturing issue.  The power system of the PCB has been doubled in capacity to ease the bottleneck caused by the need for power from the chips.  This is what we've been working on over the last two weeks. 

From what I understand, some of the additional power consumption is because of the power circuit being run above design levels. Once the power circuitry is oversized, the efficiency should increase, and the total consumption should decrease overall. Is this correct?

Correct.

[BFL]

How long before the redesigned products will be shipped? Thanks.

It depends on the purchase date.  Pre-ordered units will be delivered early January.  Delivery times on most orders will fall within the originally scheduled windows.

[kronosvl]

From what I understand the product tested by Inaba (thanks Inaba) is not the redesigned product. If that's the case when we will have some numbers with the new product (MH/s, power consumption)?
I'm asking this because from the initially 50MH/w the product is now at ~10MH/w and I'm hopping for a little increase after this redesign.

Speaking of power consumption, can the client run on a router with a custom firmware(linux based like openWRT, Oleg, or others) and USB port instead of a PC?

Thanks in advance for answering my questions

[bulanula]

Advantages to 5970s :

-less noise
-less heat
-less power consumption ( don't really care as most miners got free electricity )
-less expensive ( 600 USD is about £384 and here in the UK the damn 5970s cost £400 at least - $300 for a 5970 was a miracle over in the states )

Disadvantages to 5970s :

-same damn hashrate
-cannot resell at all
-no gaming
-no other purpose at all ( you can hack WPA on GPUs as well as mining SolidCoins )

Conclusion :

For me at least, these units are worthless right now. I would rather get me a 5970 that has many uses and some resale value rather than these units. The 5970 is noisy and makes heat and consumes a ton of power but the price is OK and the performance is the same.

What chips are you using inside ? Altera Hardcopy ? Why are you refusing to tell us that very important information ? If it is to prevent copies than we all know how easily it is to probe the chips via JTAG if you bothered to erase the chip numbers off the top.

IMHO this is still a long way off the promised 1050 Mhash/s, 20W and $499 and those power brick issues are not funny at all.
 

[fred0]

-less power consumption ( don't really care as most miners got free electricity )

That's right, because they scam for their electricity, and then spend their bitcoin on silk road.

[spiccioli]

Advantages to 5970s :

-less noise
-less heat
-less power consumption ( don't really care as most miners got free electricity )

bulanula,

it's all about power consumption.

Here where I live I pay around 0.25 EUR / kWh, a 5970 consumes a lot more than it can mine.

So, while a 5970 could be resold, after one year of 24x7 mining can it be resold at all?

My only doubt right now is about mining 24x7 with the bitforce since it has been designed to work using 1/4th of the energy it is using under real work conditions.

How long before some component on the bitforce, apart from the new power box, dies of over heating/current ?

Best regards.

spiccioli.

[Hawkix]

-less power consumption ( don't really care as most miners got free electricity )

There is no such thing as free lunch .. nor electricity.

[CubedRoot]

-less power consumption ( don't really care as most miners got free electricity )

There is no such thing as free lunch .. nor electricity.

Well, I know several people whom have electricity built into their monthly lease cost at a flat rate. So, for them electricity is "free", as they can pretty much use as much as they want for the same monthly rent on their apartment.

[DeathAndTaxes]

I think we need some fingerprint analysis on the digit reflected in the kill-a-watt, to be entirely sure that aliens were not involved.

No we don't 820MH on 82W is horrible.  Nobody would fake that.  That's 10MH/W.  Other FPGA boards are getting >2x that.

So finally nearly 2 months later after all the speculation on their "insane performance breakthrough" it turns out there is no breakthrough.  Performance per watt is inferior to existing designs which have already shipped.  It was the performance per watt claim which was so implausible.  Well beyond what current gen FPGA are capable of.  The only tech capable of that level of performance (sASICS, etc) have such high upfront costs is seemed implausible that a company from nowhere would have that kind of capital.  The reality was simpler ... It was never possible.


While needing to throttle the top speed due to power/thermal issues is understandable getting power consumption wrong by >400% isn't. I mean this isn't like saying 20W and it turns out to be 23W. It would be like Toyota claiming a Prius gets 130MPG and then at launch it turns out it gets worse gas mileage than every existing hybrid already on the market.

Generally it is a good idea of benchmark your own product before selling it w/ a specific level of performance claimed.

BTW where did the very exact performance numbers come from.  1.05GH.  the .05 implies a high level of confidence, and thus testing.  When did you see 1.05GH in the lab?  Ever?  Same thing w/ power.  The claims wasn't ~20W it was 19.8W.  You were that sure it was .8 not .7 or .9 to use the significant digit.  Did you EVER see 19.8W in the lab?

[DeathAndTaxes]

You already redesigned it? What modification? Just the power box change? I'm lost...

The external PSU is not a manufacturing issue.  The power system of the PCB has been doubled in capacity to ease the bottleneck caused by the need for power from the chips.  This is what we've been working on over the last two weeks.  

From what I understand, some of the additional power consumption is because of the power circuit being run above design levels. Once the power circuitry is oversized, the efficiency should increase, and the total consumption should decrease overall. Is this correct?

Correct.

By what factor.  I mean it seems improbable that power circuitry is THAT inefficient.  I mean say the 12V DC to 1.1V DC switching is 20% inefficient cutting that to 10% will improve power consumption marginally but isn't going to close the gap between 82W observed and 19.8W claimed. Right?

It might help w/ temps and allow chips to be pushed higher but there is something else making the power consumption 300% (on MH per W basis) compared to Spartan LX150 chips.  Last generation FPGA (65nm)?

[ngzhang]

 It was never possible. 

and by there "excellent" marketing communications, nearly all other unfirm FPGA developers scared and canceled their project.
good job.

[DeathAndTaxes]

Again, at these relatively low power levels, a board that is twice as expensive is going to take a LONG time to recover the costs.  If as you say Nzhang's boards are about 20 watts, you only have about 60 watts difference. How long does it take you to recover the additional $600 in expenses on a 60 watt difference?
Power in my area is .$09 per killowatt hour.
.095 / 1000 watts = $.000095 per watt hour.  
20 watts running 730 hours (I figured that for a month) at my rate of $.095 = ~ 14.06 KW/hours or total cost of about $1.34 / month
80 watts running 730 hours (I figured that for a month) at my rate of $.095 = ~ 51.10 KW/hours or total cost of about $4.85 / month

So that means it would take you about 170 months to recover the cost differential between the BFL unit and the Nzhang unit....thats assuming they output close the same performance, and we are only figuring power differential.

If my math is off, please let me know...its 2am and I have been drinking a little :O

No you are right, apples to apples the board "as tested" is very competitive despite the higher power consumption.  

The issue is the competence of the company.  Will it last the 12+ months necessary to just cover the cost of the capital.  What FPGA is being used?  What thermal load is it rated for?  Are the other components on the board capable of handling the 400%+ increase in power compared to what the company thought/claimed?  Does the board perform as claimed for longer periods of time?  A 24 hour test would be nice? A 1000 hour continual run test would be even nicer.  I mean lets call a spade a spade.  If a company gets power consumption wrong not by 10% or even 50% but by >400% what else did they get wrong? 

Also other boards are available w/ bulk discounts.  Also some people have power costs as high as $0.30 per kWh. So it isn't just a black and white.  The largest issue is risk/cashflow.  Mining revenue will always trend towards lowest cost producers.   Having the most efficient board is less risky.  If you have high efficiency and low power costs it is unlikely price/difficulty will ever put you in a unprofitable situation.  Granted you are accepting a lower return on capital but doing so with less risk.

[BFL]

Just to clarify...   The demo session ran well over an hour straight in Inaba's DC at the stated performance figures.  The chips themselves run at aprox 62C at this speed.  Furthermore, the demo unit has a tiny power system which does get hot because it's over burdened where the production units have a much larger power system which will run considerably cooler since it's operating within it's designed capacity.

Regards,
BFL

[BFL]

It's also interesting to note that the only negative comments are a virtual role call of the people who've staked their reputation on this product not existing at all.  I don't want to pick a fight, but at some point you may want to consider the long term implications of our company not being a scam.

[DeathAndTaxes]

It's also interesting to note that the only negative comments are a virtual role call of the people who've staked their reputation on this product not existing at all.  I don't want to pick a fight, but at some point you may want to consider the long term implications of our company not being a scam.

The product doesn't exist.  The product debated was 1.05GH @ 19.8W for $500.  That product doesn't exist. 

You have another product which seems very much real and at much more realistic performance metrics.  I don't want to pick a fight but the only one who made a false claim was you.  You "guaranteed" delivery of a product in 4-6 weeks (long since passed) that would achieve 1.05GH/s @ 19.8W for $500.  A guarantee your very young company has already failed to live up to and likely never will.

[BFL]

It's also interesting to note that the only negative comments are a virtual role call of the people who've staked their reputation on this product not existing at all.  I don't want to pick a fight, but at some point you may want to consider the long term implications of our company not being a scam.

The product doesn't exist.  The product debated was 1.05GH @ 19.8W for $500.  That product doesn't exist. 

You have another product which seems very much real and at much more realistic performance metrics.  I don't want to pick a fight but the only one who made a false claim was you.  You "guaranteed" delivery of a product in 4-6 weeks (long since passed) that would achieve 1.05GH/s @ 19.8W for $500.  A guarantee your very young company has already failed to live up to and likely never will.

The guarantee was that should we not deliver at those specifications, the order could be canceled or the final resulting performance figures could be accepted.  We have kept to our guarantee.