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

[bulanula]

So when will these ship now ? Presumably in 2012 then, right ? When is that final version of the hardware with the fixed power issues coming ?

[defxor]

Let's put this in perspective.

They were off by a factor of 3.

Imagine someone saying "I overclocked my CPU to get 12 GHz".  When they actually test the CPU, it gets 4 GHz.

Much closer to reality:

Company says "I have something that runs programs at 12 megaops"

You tell them there is a market for something that runs programs fast.

Company says they'll give you that 12 mops system they have, and then finds out that your programs have quite specific requirements which makes the product only run them at 4 megaops (while still able to execute regular programs at 12)

FPGAs aren't CPUs, nor are they GPUs, which invalidates a few of the later posts on how "off" the estimations were. Should BFL have kept quiet about expected power usage for typical scenarios and waited until they knew exactly how Bitcoin mining would task the system?

Sure, Maybe, No. Take your pick.

[Epoch]

Let's put this in perspective.

They were off by a factor of 3.

Imagine someone saying "I overclocked my CPU to get 12 GHz".  When they actually test the CPU, it gets 4 GHz.

Much closer to reality:

Company says "I have something that runs programs at 12 megaops"

You tell them there is a market for something that runs programs fast.

Company says they'll give you that 12 mops system they have, and then finds out that your programs have quite specific requirements which makes the product only run them at 4 megaops (while still able to execute regular programs at 12)

FPGAs aren't CPUs, nor are they GPUs, which invalidates a few of the later posts on how "off" the estimations were. Should BFL have kept quiet about expected power usage for typical scenarios and waited until they knew exactly how Bitcoin mining would task the system?

Sure, Maybe, No. Take your pick.

Even closer to reality:
- company says "we have a 1.05GHash/s, 20W box"
- you tell them there is a market for it (bitcoin mining) that they may not have first realized
- company says they'll give you that 1.05GHash/s, 20W box for it, but later finds out that their FPGA internal switching rate is so much higher for bitcoin mining (which they didn't target originally) that they have to underclock their system by 15% (down to 832MHash/s) and draw 4x the power (80W).

I don't think there is much doubt that such a box exists (Inaba has seen 2 in operation); BFL stated they would likely get the power draw down a bit by doubling the capacity of the onboard power regulators and by beefing up the power brick.

At this point a more relevant question would be: how does the 4x power draw affect the FPGAs? BFL's original cooling solution (small heatsinks and 80mm fan) was designed for a 20W heat envelope. Now the heat is 80W ... is that same cooling subsystem still sufficient, or will the FPGA's run too hot and have reduced lifetime? Typical designs are overengineered for 2x the nominal spec. 2x 20W is 40W; 80W is beyond even that.

[makomk]

Even closer to reality:
- company says "we have a 1.05GHash/s, 20W box"
- you tell them there is a market for it (bitcoin mining) that they may not have first realized
- company says they'll give you that 1.05GHash/s, 20W box for it, but later finds out that their FPGA internal switching rate is so much higher for bitcoin mining (which they didn't target originally) that they have to underclock their system by 15% (down to 832MHash/s) and draw 4x the power (80W).

That's not quite right; as I understand it the reason the internal switching rate is so high is the 1.05 GHash/s of SHA-256 in itself. In theory it's inherent to this kind of high-speed hashing. (Not that there's really many uses for this sort of hash rate other than Bitcoin mining anyway...)

[legolouman]

Let's put this in perspective.

They were off by a factor of 3.

Imagine someone saying "I overclocked my CPU to get 12 GHz".  When they actually test the CPU, it gets 4 GHz.

Much closer to reality:

Company says "I have something that runs programs at 12 megaops"

You tell them there is a market for something that runs programs fast.

Company says they'll give you that 12 mops system they have, and then finds out that your programs have quite specific requirements which makes the product only run them at 4 megaops (while still able to execute regular programs at 12)

FPGAs aren't CPUs, nor are they GPUs, which invalidates a few of the later posts on how "off" the estimations were. Should BFL have kept quiet about expected power usage for typical scenarios and waited until they knew exactly how Bitcoin mining would task the system?

Sure, Maybe, No. Take your pick.

Even closer to reality:
- company says "we have a 1.05GHash/s, 20W box"
- you tell them there is a market for it (bitcoin mining) that they may not have first realized
- company says they'll give you that 1.05GHash/s, 20W box for it, but later finds out that their FPGA internal switching rate is so much higher for bitcoin mining (which they didn't target originally) that they have to underclock their system by 15% (down to 832MHash/s) and draw 4x the power (80W).

I don't think there is much doubt that such a box exists (Inaba has seen 2 in operation); BFL stated they would likely get the power draw down a bit by doubling the capacity of the onboard power regulators and by beefing up the power brick.

At this point a more relevant question would be: how does the 4x power draw affect the FPGAs? BFL's original cooling solution (small heatsinks and 80mm fan) was designed for a 20W heat envelope. Now the heat is 80W ... is that same cooling subsystem still sufficient, or will the FPGA's run too hot and have reduced lifetime? Typical designs are overengineered for 2x the nominal spec. 2x 20W is 40W; 80W is beyond even that.

But this box was designed for Bitcoin. Hence the 'BitForce' Line

[finway]

It's real ?
Has anyone tested it?

[Inaba]

OMG. Seriously?

[Raoul Duke]

It's real ?
Has anyone tested it?

Can't read?

[grue]

But this box was designed for Bitcoin. Hence the 'BitForce' Line

implying they make any other boxes other than "bitforce"

[rph]

Are these shipping yet? If not, what's the ship date?

-rph

[BFL]

Are these shipping yet? If not, what's the ship date?

-rph

We're currently in manufacture with our revision B PCB with the larger power regulator.  Our expected ship date for pre-ordered BitForce Singles is mid January. 

Regards,
BFL

[RandyFolds]

Are these shipping yet? If not, what's the ship date?

-rph

We're currently in manufacture with our revision B PCB with the larger power regulator.  Our expected ship date for pre-ordered BitForce Singles is mid January. 

Regards,
BFL

So in other words, 4-6 weeks? 

[wndrbr3d]

We're sorry for the hiccup, so until the greater performance potential is demonstrated, our BitForce Single will be reduced to $599. 

I'm late to the party and am still catching up on this thread, but this line right here is suuuuuper telling on how BFL might operate as a business.

Obviously they had to know ahead of this test how their product performs, and not until a test was published do they come clean with the REAL numbers? 

That being said, I agree with other people in this thread that, if anything, this entire demonstration really makes the case for small time FPGA developers here due to their openness and honesty when it comes to their products performance as well as build quality.

Wow.

[ursa]

What i don't understand is how they came up with those figures of 20W and ~1GH/s...You have to test to find these figures...They have tested their "technology"  to "public" only ~1 week ago ,getting ~800Mh/s and 80W.(in front of a forum member) ,but i believe that their "real world test" wasn't made too long before that meeting with the forum member.
There is a big difference between theory and practice,especially in electronics...You have to test first and then advertise ,not read some specs and start advertising and collecting money.
In my opinion,BFL has little experience in this things...I am not saying that i know too much,but i have learned that theory is not = practice,especially in elecronics.Anyone who has something to do with electronics knows this.
Or they don't know shit about electronics and they had an ideea and they know person X and person Y and try to put a device together or there is a big lie around...To me ,this whole thing doesn't make sense, I mean the course of events....
Please correct me if i am wrong,but don't start to tell stupid things about this post...this is just my personal opinion.

[plastic.elastic]

What i don't understand is how they came up with those figures of 20W and ~1GH/s...You have to test to find these figures...They have tested their "technology"  to "public" only ~1 week ago ,getting ~800Mh/s and 80W.(in front of a forum member) ,but i believe that their "real world test" wasn't made too long before that meeting with the forum member.
There is a big difference between theory and practice,especially in electronics...You have to test first and then advertise ,not read some specs and start advertising and collecting money.
In my opinion,BFL has little experience in this things...I am not saying that i know too much,but i have learned that theory is not = practice,especially in elecronics.Anyone who has something to do with electronics knows this.
Or they don't know shit about electronics and they had an ideea and they know person X and person Y and try to put a device together or there is a big lie around...To me ,this whole thing doesn't make sense, I mean the course of events....
Please correct me if i am wrong,but don't start to tell stupid things about this post...this is just my personal opinion.

SSHHHHH

Sheep dont want to ask any questions.

[Syke]

What i don't understand is how they came up with those figures of 20W and ~1GH/s...You have to test to find these figures...

Those were made-up numbers meant to 1. garner attention, and 2. convince people not to buy competing products. Obvious bait-and-switch scam.

[grue]

What i don't understand is how they came up with those figures of 20W and ~1GH/s...You have to test to find these figures...

they're basically using all this forum attention to sell their mediocre FPGAs, which by the time they ship it (if ever), would probably have the same performance as existing ones. even if everyone canceled their preorders because the specs don't match, they essentially got 2+ months of interest free loan.

[Keefe]

Are these shipping yet? If not, what's the ship date?

-rph

We're currently in manufacture with our revision B PCB with the larger power regulator.  Our expected ship date for pre-ordered BitForce Singles is mid January. 

Regards,
BFL

Do you have many completed or almost completed Rev A units that you'd consider selling with a drastically reduced clock to keep the heat and power draw within spec, with a deep discount to correspond with the low hashrate? I wonder if you might have boards that are now otherwise useless because of the discovery of higher power draw than expected.

[worldinacoin]

I really hope for some production units rather than pre-order again and again

[Inaba]

they're basically using all this forum attention to sell their mediocre FPGAs, which by the time they ship it (if ever), would probably have the same performance as existing ones. even if everyone canceled their preorders because the specs don't match, they essentially got 2+ months of interest free loan.

First of all, I don't see any competing FPGAs with the same performance envelope.  Second, if by some chance someone does have a competing FPGA around the same envelope by that time, then nothing was really lost as far as "bait-and-switch" goes, since currently, there is nothing released that is even close to the neutered version that was demonstrated... and if they manage to get something that performs better than what was demonstrated, that's just icing on the cake.