[VIDEO] BFL Bitcoin ASIC Teardown

[smoothie]

http://www.youtube.com/watch?v=ciwfCqP8AiU

[computerparts]

I know people love to bitch and moan at BFL and rightfully so in some cases. But I have to say if BFL brought this chip from design to market in one year then that is outstanding. Regardless of what they promised, what they accomplished is no easy task. A lot of people aren't aware of the complications involved when attempting something like this. It can take a lot longer than a year to bring a new design to market. Avalon uses a 110nm process so it's no surprise they were able to deliver first. Technologically BFL's 65nm process is years ahead of Avalon. Remember I'm not speaking of what delivery time frame BFL originally promised, that has no relevance here. I'm speaking strictly of what they accomplished and to put it lightly it is amazing.   

[Vagnavs]

I know people love to bitch and moan at BFL and rightfully so in some cases. But I have to say if BFL brought this chip from design to market in one year then that is outstanding. Regardless of what they promised, what they accomplished is no easy task. A lot of people aren't aware of the complications involved when attempting something like this. It can take a lot longer than a year to bring a new design to market. Avalon uses a 110nm process so it's no surprise they were able to deliver first. Technologically BFL's 65nm process is years ahead of Avalon. Remember I'm not speaking of what delivery time frame BFL originally promised, that has no relevance here. I'm speaking strictly of what they accomplished and to put it lightly it is amazing.   

yeah, great when is my unit coming? Before the to end the year or after.. It may be great chip, but it will be worthless if they don't ship them out soon.

[jspielberg]

What would be great is if they started selling the bare chip and the tech specs.

They might not be able to put together a machine in record time, but if they made the chip available... I bet the response would be extremely positive.   I think everyone agrees that their chip is the most advanced one that is "available" right now (I am leaving AsicMiner out if this discussion).

All that being said, in the bitcoin mining arms race, technical elegance takes a back seat to deployability... and right now avalon's technology (while technically inferior) is by far sitting up front.  BFL's customers are looking at a tidal wave of Avalon generated difficulty which will just get worse on a month to month basis through the rest of the year kicking into high gear when their chip sales get turned into DIY hashing machines.

[wsoei]

I"m impressed at the advancement in technology and saddened by the slow deployment.

[johnyj]

The original design is 8 chips for 60GH, so each chip is designed to be run at 8GH. Now in Jalapeno there are 2 chips, but only made it 5GH, I suppose that some of the low quality chips went into Jalapeno

[johnyj]

What would be great is if they started selling the bare chip and the tech specs.

They might not be able to put together a machine in record time, but if they made the chip available... I bet the response would be extremely positive.   I think everyone agrees that their chip is the most advanced one that is "available" right now (I am leaving AsicMiner out if this discussion).

All that being said, in the bitcoin mining arms race, technical elegance takes a back seat to deployability... and right now avalon's technology (while technically inferior) is by far sitting up front.  BFL's customers are looking at a tidal wave of Avalon generated difficulty which will just get worse on a month to month basis through the rest of the year kicking into high gear when their chip sales get turned into DIY hashing machines.

True, the mining device industry is almost a suicide business, the more you deliver, the less attractive your product will become, due to higher difficulty caused by your own delivery

[JCviggen]

The original design is 8 chips for 60GH, so each chip is designed to be run at 8GH. Now in Jalapeno there are 2 chips, but only made it 5GH, I suppose that some of the low quality chips went into Jalapeno

I don't think BFL knows which chips are good and which are poor until they solder them onto a board.

[Frizz23]

True, the mining device industry is almost a suicide business, the more you deliver, the less attractive your product will become, due to higher difficulty caused by your own delivery

Given the endless trail of I-ordered-a-Jalapeno-in-April-2013-when-will-I-be-rich noobs/fools it's not a suicide business for BFL, but a suicide business for those newbie miners.

[johnyj]

True, the mining device industry is almost a suicide business, the more you deliver, the less attractive your product will become, due to higher difficulty caused by your own delivery

Given the endless trail of I-ordered-a-Jalapeno-in-April-2013-when-will-I-be-rich noobs/fools it's not a suicide business for BFL, but a suicide business for those newbie miners.

Yes, maybe that is the right move, just like Avalon, raise the price and reduce the delivery speed, delay the difficulty rise and make their life better. I think they could form an alliance like OPEC

[aTg]

We must congratulate them for completing the development of its new chip probably of more than a million dollars, it's a shame they have spent a few cents on a thermal pad instead of using thermal grease.

[wrenchmonkey]

The original design is 8 chips for 60GH, so each chip is designed to be run at 8GH. Now in Jalapeno there are 2 chips, but only made it 5GH, I suppose that some of the low quality chips went into Jalapeno

I don't think BFL knows which chips are good and which are poor until they solder them onto a board.

That may be true, but I would think that they could easily create some pogo pin testing stations, which would allow quick benchmark testing on each chip in a matter of a 5-10 seconds per device. It's not uncommon to do this prior to assembly, so that you don't waste time soldering in a DOA chip. A few seconds on the front end saves an hour of work on the back end repairing a board with a faulty chip. 

[JCviggen]

The original design is 8 chips for 60GH, so each chip is designed to be run at 8GH. Now in Jalapeno there are 2 chips, but only made it 5GH, I suppose that some of the low quality chips went into Jalapeno

I don't think BFL knows which chips are good and which are poor until they solder them onto a board.

That may be true, but I would think that they could easily create some pogo pin testing stations, which would allow quick benchmark testing on each chip in a matter of a 5-10 seconds per device. It's not uncommon to do this prior to assembly, so that you don't waste time soldering in a DOA chip. A few seconds on the front end saves an hour of work on the back end repairing a board with a faulty chip.  

Didn't one of the guys who posted about receiving a Jalapeno get one with lower hashing power because one chip turned out to be a dud so they disabled it and clocked the other one higher but falling short of target?
Nothing I've seen from BFL indicates they have any clue about their chips and simply solder them on and see what happens. Gives you a bit of a clue how "streamlined" their production process is at this stage. Hence the very low volume.

[r3animation]

The original design is 8 chips for 60GH, so each chip is designed to be run at 8GH. Now in Jalapeno there are 2 chips, but only made it 5GH, I suppose that some of the low quality chips went into Jalapeno

I don't think BFL knows which chips are good and which are poor until they solder them onto a board.

That may be true, but I would think that they could easily create some pogo pin testing stations, which would allow quick benchmark testing on each chip in a matter of a 5-10 seconds per device. It's not uncommon to do this prior to assembly, so that you don't waste time soldering in a DOA chip. A few seconds on the front end saves an hour of work on the back end repairing a board with a faulty chip.  

Didn't one of the guys who posted about receiving a Jalapeno get one with lower hashing power because one chip turned out to be a dud so they disabled it and clocked the other one higher but falling short of target?
Nothing I've seen from BFL indicates they have any clue about their chips and simply solder them on and see what happens. Gives you a bit of a clue how "streamlined" their production process is at this stage. Hence the very low volume.

I believe they had to use 2 chips because if they had one on the PCB the heatsink wouldn't sit probably and many bad things would happen.

[JCviggen]

Pretty sure I read that BFL themselves told the dev who got it that they had disabled a faulty chip and clocked the other higher. Ended up doing 4 GH/s. Other people have received Jalapenos with both chips functioning and doing the advertised hashrate of 5 or better.

edit: https://forums.butterflylabs.com/hardware-discussion/1983-grnbrgs-unboxing-teardown-new-bitforce-5-gh-s-sc-miner.html

[r3animation]

Pretty sure I read that BFL themselves told the dev who got it that they had disabled a faulty chip and clocked the other higher. Ended up doing 4 GH/s. Other people have received Jalapenos with both chips functioning and doing the advertised hashrate of 5 or better.

edit: https://forums.butterflylabs.com/hardware-discussion/1983-grnbrgs-unboxing-teardown-new-bitforce-5-gh-s-sc-miner.html

Yea I know. I'm just saying they couldn't ship him a 1 chip miner as the hsf would be un-balanced.

[JCviggen]

Pretty sure I read that BFL themselves told the dev who got it that they had disabled a faulty chip and clocked the other higher. Ended up doing 4 GH/s. Other people have received Jalapenos with both chips functioning and doing the advertised hashrate of 5 or better.

edit: https://forums.butterflylabs.com/hardware-discussion/1983-grnbrgs-unboxing-teardown-new-bitforce-5-gh-s-sc-miner.html

Yea I know. I'm just saying they couldn't ship him a 1 chip miner as the hsf would be un-balanced.

Obviously, but given the fact that a regular Jalapeno has 2 functioning chips and this one doesn't...they would either have to have no good chips left to use or have soldered on the dud and found out it didn't work only after that. It didn't appear to be an intentional move to create an inferior Jalapeno and ship that out. In fact, I don't really understand why they shipped that one at all. Why not spend a few minutes (based on the 400 units a day lol) to make a properly functioning one and ship that instead? They would've spent time changing the firmware on the semi dud one anyway.
It's pretty clear that they only discovered the issue when they powered it on the first time. Likely conclusion, their chips are not binned or tested.