Hacking Bitmain Antminers (S7 & S9) because man a lot of these break......

[lightfoot]

So although I have steered clear of Bitmain miners in the past I figured enough people have been having issues with these miners and it's time to figure out what the heck is going wrong with them. So I'm starting yet another thread on hacking these things to either fix them or at least figure out why they blow up.

This will be under my normal license: Information should be free, skills should be paid for. Therefore as I find out stuff about these miners I'll put it here. If anyone else has thoughts or technical specs please feel free to share them.

I'd rather this not turn into a general discussion on the merits of Bitmain, Bitcoin Crash, or anything else like that, will try to keep it to the technical bits.

The second post will be a FAQ of most interesting stuff, and the third post will be reserved for other administrivia.

On with the show. Or as I like to say:

"What's the worst that could happen?"

[1713630842]

1713630842

[1713630842]

1713630842

[lightfoot]

Reserved for FAQs.

[lightfoot]

Reserved for administrivia

[lightfoot]

Ok, let's get started. Our latest board of the hour came in with the usual it doesn't work problem. There are a couple of types of it doesn't work:

It blows up the power supply and shorts out.
It comes up but does not hash
It doesn't come up.

First thing to check is to see if anything is unusual about the board. In this case, not really other than the fact that the heat sinks look like they were put on by a 5 year old. I spend a lot of time aligning heat sinks on S7's to be perfectly straight to minimize airflow turbulence since these things are air cooled in small boxes.

However I did notice that one was a little bit loose, could wiggle it a bit like a loose tooth. That's bad. Using the nose I could smell a bit of burned smell at that point on the board, greeeeeat......

So put the board on the preheater, warmed up the board, then used the air tools to warm up the heat sink. Not too difficult, and it came right off leaving this:



Now for a close up of the chip...


And a side view of the chip.


And a view of the heat sink itself.


Using these we can see the problem: This chips is one of the ones that does not have a heat sink on the bottom, and more importantly has only about 50% of the chip covered with the heat sink glue.

From what I can see here it looks like a sloppy Bitmain job when building it led to a board that would run somewhat warm anyway. The chip itself only had 50% contact with the sink, and judging how thick that compound is it looks like the compound made a poor connection between the chip and sink. Over time the chip got warm, had a lower resistance because it is in a series string, pulled more current, more heat, and the usual failure.

Bitmain should warranty this board, but it's probably out of warranty. The lack of compound on the whole chip points to a manufacturing fail.

Solution: We could pull the chip, but that gets complex and I need to talk to the owner before doing that. Likewise if the SCL signal goes through every chip in the board in series, removing the chip breaks the chain and the board does not work. Drat. Wonder if bitmain will sell me a bunch of chips....

[Isaac_]

You can order those chips on alibaba. Perhaps even aliexpress.

Could you replace said chip?

If so, just replace, add thermal paste and heatsink, should be good to go.

Pardon if my suggestion isn't the best, just trying to help. Know a good deal about computer hardware but this isn't my expertise.

[barmitzfit]

You can order those chips on alibaba. Perhaps even aliexpress.

Could you replace said chip?

If so, just replace, add thermal paste and heatsink, should be good to go.

Pardon if my suggestion isn't the best, just trying to help. Know a good deal about computer hardware but this isn't my expertise.

I don't think it's as simple as that. I have been researching buying ASIC chips from alibaba and noone has yet to even return a pinout diagram..

Looking at some other threads here people have been trying to find a ASIC distributor but BitFury seems to be the only one coming out with a consumer 16nm ASIC.

[Isaac_]

I could find many sellers for the bitmain chip. I asked for a bulk amount, which was about $8. I'm guessing just a few chips will be more expensive per unit, or they might want a minimum order of some amount (I hope no higher than 10).

As for pinout diagrams, it seems they would be your only real problem, shame I can't help you with that. I am pretty sure you can buy these chips though. At least for the S9. I don't know about the S7.

-edit-

Ah, it's an S7. I would have looked up for you on alibaba, but I can't find the specific chip? Ah, found it. It's a BM1385. Yes, one seller does seem to have it for sale. https://www.alibaba.com/product-detail/Miner-ASIC-Chips-BM1382-BM1385_60707278004.html
Try contacting them. Hope it's any help. Good luck. Would help you more as barmitzfit said with the pinout chart even if you can buy the chips it seems there will be more problems.

There seems to be many more retailers for the S9 chip than the S7

[lrowland21093]

I found the datasheet for the BM1385 chip.  I am a newbie so I probably cannot post a link (though I have been reading and learning here for years).
I will try anyway...

https://bits.media/images/asic-miner-antminer-s7/BM1385_Datasheet_v2.0.pdf

[barmitzfit]

I could find many sellers for the bitmain chip. I asked for a bulk amount, which was about $8. I'm guessing just a few chips will be more expensive per unit, or they might want a minimum order of some amount (I hope no higher than 10).

As for pinout diagrams, it seems they would be your only real problem, shame I can't help you with that. I am pretty sure you can buy these chips though. At least for the S9. I don't know about the S7.

-edit-

Ah, it's an S7. I would have looked up for you on alibaba, but I can't find the specific chip? Ah, found it. It's a BM1385. Yes, one seller does seem to have it for sale. https://www.alibaba.com/product-detail/Miner-ASIC-Chips-BM1382-BM1385_60707278004.html
Try contacting them. Hope it's any help. Good luck. Would help you more as barmitzfit said with the pinout chart even if you can buy the chips it seems there will be more problems.

There seems to be many more retailers for the S9 chip than the S7

The only reason I asked for a pinout was to determine if they were a scam or not. I mean, how can a company that sells these chips not have any technical information? I don't think the ASICs on alibaba are real, I see a lot of threads on here discussing how people can create their own SHA256 ASIC and most of the answers come back saying it's impossible (The equipment used to make them is in the millions).

edit: I mean this from my own expierence, I contacted a few of these suppliers and they never answered after I asked for any technical documents

[lightfoot]

It's always tricky to find hashing chips, and normally it is not the chips themselves but the supporting technology that goes foom. Usually it's either the clock crystal (S7), the power scale chip (s7) or the FETs (S7, S9) that die, Those are easy to replace.

When hashing chips go bad it normally sucks: There's possibly damage under the chip and although these are not as bad a QFN chips they are still a pain and finding a blown chip under these heat sinks is not a joy. On Avalon units the solder typically extrudes under the chip, shorting out the communication lines. On a unit with individual heat sinks like Ants, it's normally the chip that burns up.

This is also a S9, not an S7. There are plenty of S7 chips, just grab a torched board and pull chips off it then re-tin and swap.

C

[dlezama]

Interesting stuff, I'll keep an eye on your progress! I guess an option is to start buying broken stuff and cannibalize spare parts. Some people around here have many broken boards, I guess they may be happy if you can get them a few Frankenstein miners out of their garbage

[score1more]

Ok, let's get started. Our latest board of the hour came in with the usual it doesn't work problem. There are a couple of types of it doesn't work:

It blows up the power supply and shorts out.

Is there a solution that could be developed to protect the power supply?

[lightfoot]

That's an interesting question: Normally when you have a boost/buck power supply you want to size your fets for both current capacity and dwell time. Dwell time is the time the FET has to spend "on" and the amount of time it takes to turn it "off" then "on" again.

So for example if you have to step 12v down to 1v for a single chip miner you need FETs that are on 11/12th of the time to ground, and 1/12th of the time to +12. This means that the high side FETs have to switch 12 times faster than the low side ones. The transition time is when the heat occurs, because you're switching and you want that period to be as short as possible. Likewise when you turn off you have to have the gate interrupt the power flow, and that takes time/generates heat. So you want a really fast switching FET with a narrow trench and low capacitance.

Problem is the best high current FETs have nice large gate junctions so they're slow. You can switch at a slower rate, but that causes a lot more ripple which has to be smoothed by the chokes and the capacitors. And THAT generates heat.

This is why on the S7's they used two FETs on the high side (12v) and one FET to switch the low side. Since they were only bucking from 12v to 10.5 or so most of the time the FETs are "on" with a small amount of "off". For the S9's they seem to start at 9.5v or so, and thus added a second FET on the low side. Which might be causing problems including more heat (thus they stuck a heat sink on the back side of the FETs, surprise!) as well as more shorted FETs which short out the power supply.

They should have switched to a different single FET which would have lower capacitance (capacitors in parallel add value like resistors in series) but for some reason they didn't.

By the way, this seriously nailed BFL on the Monarchs. They finally went with high power high frequency FETs on the high side but you can see where they had the pads to run TO style FETs. They also had 6 phase power supplies to smooth out the current, but that meant each FET was switching 6 times as quickly. Real bitch on wheels, if they had put their engines in series instead of parallel it would have been more like the Ants and easier to design.

Another side note: This is why you don't ever run an antminer at higher than 12v. Higher input voltage would just require more bucking which means more heat and the usual failures.....

I think a similar problem exists in the S9's. Might explain some of the reliability problems. Fortunately replacing FETs is not hard if you have air heat and significant preheating ability to remove the FETs and reflow new ones.

[score1more]

That's an interesting question: Normally when you have a boost/buck power supply you want to size your fets for both current capacity and dwell time. Dwell time is the time the FET has to spend "on" and the amount of time it takes to turn it "off" then "on" again.

So for example if you have to step 12v down to 1v for a single chip miner you need FETs that are on 11/12th of the time to ground, and 1/12th of the time to +12. This means that the high side FETs have to switch 12 times faster than the low side ones. The transition time is when the heat occurs, because you're switching and you want that period to be as short as possible. Likewise when you turn off you have to have the gate interrupt the power flow, and that takes time/generates heat. So you want a really fast switching FET with a narrow trench and low capacitance.

Problem is the best high current FETs have nice large gate junctions so they're slow. You can switch at a slower rate, but that causes a lot more ripple which has to be smoothed by the chokes and the capacitors. And THAT generates heat.

This is why on the S7's they used two FETs on the high side (12v) and one FET to switch the low side. Since they were only bucking from 12v to 10.5 or so most of the time the FETs are "on" with a small amount of "off". For the S9's they seem to start at 9.5v or so, and thus added a second FET on the low side. Which might be causing problems including more heat (thus they stuck a heat sink on the back side of the FETs, surprise!) as well as more shorted FETs which short out the power supply.

They should have switched to a different single FET which would have lower capacitance (capacitors in parallel add value like resistors in series) but for some reason they didn't.

By the way, this seriously nailed BFL on the Monarchs. They finally went with high power high frequency FETs on the high side but you can see where they had the pads to run TO style FETs. They also had 6 phase power supplies to smooth out the current, but that meant each FET was switching 6 times as quickly. Real bitch on wheels, if they had put their engines in series instead of parallel it would have been more like the Ants and easier to design.

Another side note: This is why you don't ever run an antminer at higher than 12v. Higher input voltage would just require more bucking which means more heat and the usual failures.....

I think a similar problem exists in the S9's. Might explain some of the reliability problems. Fortunately replacing FETs is not hard if you have air heat and significant preheating ability to remove the FETs and reflow new ones.

So it's not something a simple fuse could fix...  When you say air heat and preheating ability.  Are you referring to a solder rework station???   I was looking at a Hakko rework station.

[Sandal_Hat]

This means one way to improve it is to open it up put our own heatsink glue and cover it 100%? Hmm but messing with the board means the 6 month warranty is gone I guess. Noob here.
Do u think this is why some antminers are hotter/colder than others? It is due to the heatsink glue?

Hmmm, perhaps u can buy a dead hashboard from someone and salvage some chips. One dead hashboard with some chips can help fix alot of other hashboards?

[lightfoot]

Quite possibly, this board was assembled by what I like to call a shirker. I've seen other pictures on Ebay and wondered if the sinks were loose/off, but it seems they're glued on.

It does take some time to re-mount sinks properly, you need to put a thin layer of adhesive compound on the chip, equal amount on the sink, press together and hold/align, etc. They just went whomp.

As for putting sinks on the FETs, it's best to put them on the board behind the FETs. TO style chips have a nice big drain pad and are designed to pull heat *down* and not up through the gate. Putting sinks on top of the FETs pulls hte heat through the chip, you want it to go away into the board.

C

[lightfoot]

Meantime just fixed a pair of S7's that came in the door. Both had pretty burned plugs:



And after repairs we have nice new nickel plugs...



The trick to doing a proper plug replacement is as follows:
1) preheat: You have to get the board hot enough to allow the solder to be molten for a bit, both to get the old plug off and to remove the old solder.
2) rosin solder: Has a slightly lower melting point than ROHS crap: Use it to alloy the solder on the pins, reduces the temp a lot
3) Air heat: Use to heat all the pins at once allowing you to easily remove the plugs.

Then clean up the pads, flux the new plug, put in, warm up, and solder. A good preheater can make the difference. Trying to do it without one will result in damaged pads and Vias, which will not be able to conduct the proper amount of current.

[cableiso]

I could find many sellers for the bitmain chip. I asked for a bulk amount, which was about $8. I'm guessing just a few chips will be more expensive per unit, or they might want a minimum order of some amount (I hope no higher than 10).

As for pinout diagrams, it seems they would be your only real problem, shame I can't help you with that. I am pretty sure you can buy these chips though. At least for the S9. I don't know about the S7.

-edit-

Ah, it's an S7. I would have looked up for you on alibaba, but I can't find the specific chip? Ah, found it. It's a BM1385. Yes, one seller does seem to have it for sale. https://www.alibaba.com/product-detail/Miner-ASIC-Chips-BM1382-BM1385_60707278004.html
Try contacting them. Hope it's any help. Good luck. Would help you more as barmitzfit said with the pinout chart even if you can buy the chips it seems there will be more problems.

There seems to be many more retailers for the S9 chip than the S7

The only reason I asked for a pinout was to determine if they were a scam or not. I mean, how can a company that sells these chips not have any technical information? I don't think the ASICs on alibaba are real, I see a lot of threads on here discussing how people can create their own SHA256 ASIC and most of the answers come back saying it's impossible (The equipment used to make them is in the millions).

edit: I mean this from my own expierence, I contacted a few of these suppliers and they never answered after I asked for any technical documents

+1 to this guy for asking the right questions.  Who rides with Ali Baba?  The 40 thieves.

[Isaac_]

On another hand, Alibaba does have seller protection, if you pay thorugh Alibaba or Paypal. I understand your skepticism, but I am equally skeptical to your skepticism. It could go either direction.

[cableiso]

Ali express has seller protection.  Alibaba you're on your own to negotiate a payment in a safe format.