Bitcoin Forum

Hi all!
I've 10 antminer s5 and 4 s7 miners. I think mining is interesting and it's good hobby  for me.
4 antminer s5 hashboards (v1.91) failed, they went only 4 months. I bought them directly from Bitmaintech. Temperature and power is optimal in my mining room. My 20% percent breakdown ratio is very frustrating. What is you experience about s5 failure frequency? (Power and internet connection is very stable.)

I decided that I'm going to figure out how can I fix them. Of course there is no official service manual :( (the BM1384 chip pdf https://bitmaintech.com/files/download/BM1384_Datasheet_v2.1.pdf  (https://bitmaintech.com/files/download/BM1384_Datasheet_v2.1.pdf) is very short-spoken).

I know the ASIC chips are in a chain and I believe that they not fail simultaneously. I suspect if there is a problematic chip in the chain the board aren’t able to hash, so I have to find the bad ASICs and I have to shunt out somehow.
I know that on a perfect board there is 0.81 V on the big black capacitor which lays between 2 ASICs and on a failed board I measure different voltages 0.3 - 1.3 on these capacitors (eg. 0.81V on 12 and bad voltages on 4). My theory is that where is not 0.81V on the capacitor there can be te failed asics.
If somebody find and separate a bad ASIC please write about it.
I know there can be several other reason for not hashing but I believe that there are only a few typical problems which occurs the symptom which can be read on many topics (no hashing and '-' at Asic status). Some of my board there is 'o' at status but not hashing and the red led illuminates so I think that boards have another type of problem. I decided to figure out the reasons of the malfunction and find mending solutions. 
Any information could be helpful!

I've ordered a 500Mhz oscilloscope to better understand how a hashboard function, I'll write my finding about the theme.
I haven’t find any detailed descriptions about mending if there is please send me! Thanks all!

With them being a private company we likely will never know failure rate.  We know it was small enough they kept pumping them out and made a profit.  But how many died.... we don't know.

The S5 for what it's worth ran hot and loud.   I was not the biggest fan of the plastic sides, just seemed cheap compared to other designs.   But I have one that was a batch one still alive and kicking.  Mine has had hot summer, lost network, lost internet, etc.  And never has had a big issue.  Only thing was a few times it seems 1 blade goes off and it needs a reboot.

So I think it varies greatly.

You are on the right track checking the voltage across the capacitor, it will give a good clue as to which ASIC's have failed. However this is made more difficult by there being 2 Asics at each node also a high or low voltage on a faulty node will also affect the voltages on the other good nodes. Shunting is unfortunately not practical as you would also need to shunt the data signals that pass through the chips and reduce the supply voltage to allow for the shunted chips.

Rich

Someone certainly has the wiring diagram and measured the components. But that knowledge allows you to repair damaged PCB's.

There were two types of v1.91. One without small heatsinks, second with small heatsinks on both sides.
I have one damaged hashboard from the first version. Worked in ideal conditions. Dead after 5 months.

You're right. I think Bitmaintech has a detailed circuit diagram :). I've written them to send me it. If they dont send it i promise I'll do the work. It's not a difficult thing, its very schematic. I've shocked some puritan things on it (eg. the oscillators and the VRM). The problem is that I'm not a professional pcb designer and I'm not so familliar with that CPU arrangement. So every information is a great help for my mission.
I think the board is the same on both v1.91 type, only difference is the small heatsinks. I have only one board with small heatsinks from the last (secondhand) batch but its working (yet) :).

Thats the problem, a faulty ASIC turn over the voltage drop for the whole board. I'think I'll put a 4.1 V Zener paralell with the big black capacitor just only bad nodes which set up the normal voltage ratios. A bad chip will be sigend with x after a while so it makes fault only for the first 3-5 minutes and with the zener i can protect the healty chips.

I do not think a 4.1V zener is going to do anything as there is only about 0.8V / Chip. Also even if it does work you need a chip to be working to pass the data through to the chips further up the chain.

Rich

Sorry 0.8V Zener,  I dont know why wrote 4.1 :-\

The ASICs are chained such that every chip acts as a relay for all signals for the chips farther down. You get UART, flow control, and clock passing through. Each chip burns about 10W so your diode would have to be pretty stout even if signal wasn't a problem.

I'm sure they have it along with many other document's and information on it.  But chances of them opening up and actually giving it out to someone I say is very small.

At this point you pretty much can't do anything on a bitmain machine without breaking warranty.  Which I understand they want it to be them to touch and don't want people doing mod's without losing warranty.  But they have sure tightened up compared to where they once were.

In practice a circuit diagram is not going to help you much. It's all a very simple repeated string of ASICs. With the BM1384 data Sheet and the S1 Circuit which is available you are about there. Then after you have checked a few voltages the big issue is to identify which ASIC(s) have failed and salvage replacements from another S5. They are not however the easiest of chips to remove because of the heat sinking effect of the ground & power plane. You also have the additional problem of never being quite certain that the chip you have salvaged is good or not.


Rich

Yes, the principal could be the same, but It would be better if Bitmain let issue a v1.91 service guide not only the basic tutorial how to realize wich board is wrong and push the chips with you finger :).  I havent see the S1 Circuit diagram yet ,could you be so kind and send me a link about it (I'm sercing about 2 days for some miner circuit diagram without any luck :(. This will be a treasure for me cause it helps me understand this architectura. )
I can see the clock and the dates with my scope, but that is the status (I push evry chip hard with my finger, I promise :)):

https://drive.google.com/file/d/0B26YInue5gCgSzFRVjIxeDBjaHc/view?usp=sharing (https://drive.google.com/file/d/0B26YInue5gCgSzFRVjIxeDBjaHc/view?usp=sharing)

After starting evry ASIC is 'O' but makes mistakes, after a while its rezognizing the bad Asics and change the first six status for 'x' and only hash ~450ghs.
I'm wondering how have to use the test pin on the ASICs. But is think it is only usable in a sepcial test cradle. Case I feel the power in myself to solder them out, but I dont sure about the ASICs problem. I suppose that the first 2 digit is refering to the middle chips of the top level.

Here is a link to the S1 Circuit Diagram

https://github.com/AntMiner/AntGen1/tree/master/PCB

It is of course quite different to the S5 in that it is not a string design but uses Buck Converters for the Core Voltage, however if you also download the BM1380 Manual to compare with the BM1384, it does help you understand how the data signals are strung together, oscillator etc.

My limited experience is that so long as the 14V up converter is ok, the 1.8V IO voltage is ok on each chip & the oscillator is running then if the board is not hashing it's almost certainly a faulty ASIC that needs to be removed & replaced with a salvaged one.


Rich

THANK YOU!!!!!  ;D Extremly good material :)!

Maybe the ASIC are good, cause while I'm testing-restarting-measuring again and again that happened  ???:

https://drive.google.com/file/d/0B26YInue5gCgWE1seEZFVTl6NFE/view?usp=sharing

It was very interesting couse this test least at ~6 hours and I measured several time the step up voltage output after the diode, and there was randomly voltages between 12.41-14.1, I thought maybe my new scope not perfectly calibrated so I dont car about it. And before the burning I watch the PWM signal by the way and it was totally different from a perfect board. But the board was hashing with 24 chips so I left running and thinking 1 hours ago I smellt the smoke :).

I'll change the burnt parts with a step up modul (~5 USD) and I hope the ASICs are OK.

I'll drow the interesting PWM signal, because - in my opininon -  that signal form could predict the burning. (I forgat to save so I'll draw by pencil :( )

did this end up working for you? very curious.

I'll do it! But my little daughter was in hospital so i dont have time to do this job. She is ok now and we are at home  : ;D
Bitmain answer for me, but only bulshit. They wont give the circuit diagram :(.
One of my friend - who is very very good in VRM circuts will help me next week to figure out the voltage dropping principale on the chain :) so I'm very optmistic :). I'll write down the things.