Does anyone have any idea what the LEDs on the boards represent? been trying to repair some boards and they all have some variation of the LEDs . I had one board that had an incorrect feedback resistor in circuit was running about 0.7v, just one one pair of A2s. Seems to be either people work on things with hammers or I guess they get dropped a lot. Lots of sheared off components. Anyone else work on these boards and have any stories?

Looking at them now after testing I guess they relate some how to the number of chips found, maybe a binary count to a point even though when I bypass some it kinds of relates to that. Something else that is odd if I bypass any of the chips on the left chain I don't get any communication, I can do them on the right. Also the left chain seems to have a 3Mhz SPI bus and the right only has a 1.5 Mhz bus. Anyway just some more observations

As far as power it you did hook it up backwards and the psu did turn on you'd have a lot of broken stuff, like it is at the least you will have a +5 to gnd short back through the communication cable. After finally getting a complete A2 unit I have found that the engineering is just enough to get by, if you notice it is on a 8 pin header but with a 6 pin connector , probably had millions of them in stock, I just soldered screw terminals into the empty header . I have found that these units are quite annoying to keep running so far, board resets all the time , I guess time to overvolt..

EDIT

I do know the blue boards like emdje has the power is reversed I can verify that for sure

I have a few but I could be wrong, I would like to know a bit more I see that it is running with the --hwreset , do you notice if the green LEDs on all the boards go off then one cycle of red flashing then green (not when you first turn it on, every time you run cgminer command)?, I know on my A2 boards the reset pin was not connected to the MCU on board so using --hwreset to reset them does nothing, instead they need a software reset like the official innosilicon binary does, while with a small amount of soldering it is fixable if your comfortable with that. Anyway I could be wrong . I actually have a real A2 on the way not just a bunch of hashing boards I bread together into a miner, I may be way off and just adapted to what worked. Also are they green boards or blue?

I'll check it out tomorrow and see, I actually have been running mine at 1.0V here lately, definitely drops the HW errors, as for connectors I pulled them off and just put some screw headers on there and got rid of the molex plugs. I had a bad experience with those PCI-E cables a long time ago in the GPU days, sure the outside of the cable says 18 AWG and is the same diameter but inside the wire was like 24 AWG, nice Chinese cables.

EDIT:
I tired the new version it did not work with my boards, what had you changed to try and fix this?

I would be interested in seeing what your hashing boards look like, all the ones I have seen there is really no way to adjust the voltage even if there was on option to change it. going beyond that if it does change voltage the higher the better, I am running at 0.94v and 1250 works OK, it is really a trade off in heat/speed ,higher voltage is higher heat but lower errors, but at some point you may not be able to remove all that heat and you have errors anyway.

Is there any way you can post a shot of the screen with the voltage adjustments? 0.81 seems pretty low, I think stock for 1000MHz was like .88

I know a lot of people are not into taking stuff apart but if you had any info on the hashing boards like color,any version marking I would be curious to know.

I will try to help, un hooking one should not be a problem I don't think, it might be better if you could ssh into the box and run commands in a terminal to see what is going on, something you could run to test boards is run the cgminer command with the --cs option that will select boards individually I use this to test things

cgminer --stmcu 0 --spiSpeed 4 --cs x

where x is the board you want to test 0-5 or 8 selects all off them, of course you need to put the -o and the pool info in there if you want it to hash or it will detect the board and exit, normally when they have +5v and +12v they will flash green until the raspi connect to them then they will go solid. I think mine have all green and one red maybe, if anyone knows what they mean that would be cool as well, I know if they flash all red there is no +12 to them no lights at all then no +5 from the controller.

I would check power for sure and communication cables, also if you think you know which one is the problem try swapping cables and see if the problem moves or stays at that board.

I think I'll start a new thread so we can talk hardware a bit more, I do plan on releasing the code I have for the FPGA, I am trying to pretty up my setup today, got a cheap cpld board from china that is a lot smaller and mounting everything down at least.

I am using the one of your versions, from maybe v4 or the mini, to be honest I don't think the driver determines the chip count so much. Looking at the SPI traffic with the logic analyzer I see that the software asks for a chip count from the uC on the board it replies with a 0x000C which is 12 then it steps through them one at a time getting info from them, of course 6 and 12 return no data , the thing I find odd is chips 9 - 12 always return 0x14 for the temperature which makes me think they didn't modify the code on the uC on board when they went from the 8->10 chip versions or not that part of it, at least I think that is it, I know there are 4 chips that are 20C all the time.

I think a  lot of my issues in overclocking is heat, should have some heatsinks this week , they run around 55C right now I hope that will help some.

I have done a lot of thinking about more boards , should be no problem to do 8 (with 3 cs lines) on hardware but power and space is another issue, you looking at 150A+ at 12V that gets to be a lot while not impossible though.

yea got one super happy and one super sad on the same board, i have noticed that every now and then they will be very evenly spread, normally when I am tracking down something. Here is the code to my thing , not amazing but works . I got some heatsinks on the way and maybe two more cards, don't know yet.

perl bad_chips.pl /path/to/logfile


bad_chips.pl

if your saving log files or anything and like perl I wrote a little thing to parse the files and tell you which chips had errors you get some output like this, as you can see I have one really happy chip on the second board.


Board 0 Chip  1 Errors  2228
Board 0 Chip  2 Errors   767
Board 0 Chip  3 Errors   390
Board 0 Chip  4 Errors  1504
Board 0 Chip  5 Errors   408
Board 0 Chip  6 Errors     0
Board 0 Chip  7 Errors  1331
Board 0 Chip  8 Errors   775
Board 0 Chip  9 Errors   318
Board 0 Chip 10 Errors   701
Board 0 Chip 11 Errors   567
Board 0 Chip 12 Errors     0

Board 1 Chip  1 Errors  2731
Board 1 Chip  2 Errors   755
Board 1 Chip  3 Errors   935
Board 1 Chip  4 Errors   271
Board 1 Chip  5 Errors   548
Board 1 Chip  6 Errors     0
Board 1 Chip  7 Errors     2
Board 1 Chip  8 Errors  1048
Board 1 Chip  9 Errors 15264
Board 1 Chip 10 Errors   557
Board 1 Chip 11 Errors  2558
Board 1 Chip 12 Errors     0

yea there are a lot of options, those were just the HW errors when cgminer exits out on the board, yea I guess really I am talking just a few mV that isn't to bad, I may put the board back to stock and sweep the frequencys again , as for current I am using one of these http://www.alliedelec.com/fw-bell-bb-100/70103851/ works pretty well but have to get the offset and do some math but the meter takes care of that , oh yea the +-15 supply that is annoying as well .

I notice there is a thermal pad on the top heatsink, did  they use one on both sides or just the top or is that something you did?

The blue board I have has some no name dc-dc converter ICs and it looks like the feedback is something odd like a 11k and a 13k so right now I don't know the math to make that work out unless there are other parts in line else where.

I had went with single parallel option on the 10k, seemed to work fine, running at 0.94v although math says 0.92v, wish I had done some more testing before hand though but hey that is the way it is. Good job!, while crude this is a chart I did with this board, I just ran 50k shares at each frequency, I should have ran more but in the interest of not taking 3 hrs for a test.

Frequency| 1100 1120 1140 1160 1180 1200 1220 1240 1260 1280 1300 1320 1340 1360 1380 1400

HW errors| 2 2 0 4 0 5 4 1 10 9 17 14 36 43 90 89

Current on 12v| 14.28 14.45 14.67 15.06 15.23 15.46 15.64 15.76 16.13 16.37 16.63 16.75 17.2 17.44 17.53 17.89

Watts 173.502 175.5675 178.2405 182.979 184.8922 187.6844 189.8696 191.3264 195.8182 198.5681 201.7219 203.1775 208.636 211.1984 212.2883 216.6479

yea that is kind of the plan to locate bad chips, after doing a lot of testing on HW errors , the blue board has one chip that is not totally bad but far worse than any of the rest,on a test of submitting 500k shares @ 1200Mhz that board generated 59 HW errors and that chip had 46 of them.

I'd like to have more of the 12 chip boards I think I found a place over seas that has a 10 pack of chips listed for $80-$110, whether that is real or not hard to say, I only overvolted the green board went from 0.9 to 0.94v just cause I could identify what to do, luckily there was an easy way to mod it, there was an un-populated resistor place in parallel with the 10k so I put a 100k on it to give it a little more, I do know if you put a 10k in there by accident you get 1.17v which it did survive sitting idle for 30 seconds , I was just thinking of reverting them just to test before and after, I got ahead of myself on it and didn't spend a lot of time testing.

The bad thing about the blue boards are of course you need 2 more inches of heatsink if you were going to try and use the other sockets. And heatsinks are my main issue at the moment.

I am the one with the 12 chip board, I don't have any extra chips to populate it but it has everything it needs just have to mount the chips and remove the loop resistor in the chain. I am not sure but I think there adapted A1 mining boards from what I understand .

As for power supplies, I know it is not for everyone but getting away from ATX power supplies would be better and cheaper for sure, those two boards I have I removed the PCI-e power jacks and just mounted a screw terminal header, something that is odd is between the green on and the blue one the +12v polarity is different so you have to watch that.

After measuring power use on the +12 line the green board uses ~16A on the +12 line @ 1200 MHz overvolted it is about +2A more.

A question I have is does anyone know how they counted the chips, I would assume that 1 & 7 are closest to the MCU on the board, of course they report chips on 1-5 and 7-11 with 6 & 12 being 0, what I am wondering on my blue board chip 9 has a very high HW error rate compared to the others about 10x more, just wondering which one to change, which I can figure out but just wondering on some ideas.

Well the controller was easy enough to replicate in a PLD, really worked just like it should despite my best efforts to mess it up. While the FPGA I used is a million times overkill it was something I had should be able to fit it in a $0.90 CPLD

So nothing comes on? I don't really have a A2 miner but I have see photos of the controller board, I am going to assume that you do have power on the 12 and 5V rails , I don't know if the 3.3v is needed nothing that I have seen needs 3.3 just 12 and 5 and it looked like there might have been regulators on the board already for a 12->5 conversion. I would check power on the Rpi, you should be able to probe the capacitor at the normal input jack , if you got 5v there and it is still not on then you might have a big problem. On the connectors going down to the hash boards I *think* that pins 2&4 are +5 and 6,8&10 are gnd so you could check that as well. Really I think all the controller board does is bridge on the ATX psu and power the PI, I dont really know why it would need +12 at all unless it does some basic power monitoring to see if there is power there.

This are all guesses really , if you could get me a good straight on shot of front and back, I am working on a new controller board, I just only have one card I have adapted to the RPi, that power supply is pretty $$$ I hope you didn't have to pay that much for it.

What server PSU are you attempting to use? How much of it does not work, any , some lights or just dead? have you checked to see if there is output on the lines?

I have two boards at the moment a green one that says it is official(v3.0) and a blue(20140625) one that has 12 chip pads and 10 chips. What I am wondering what are peoples experiences with these boards? What I have found out so far is the green board has a HW reset ran to the 10 pin connector in pin 1 a  low->high will reset board , even though it was not connected at the connector. The blue board has the same thing but will not reset it, looking at it, for some reason it traces out to pin 15 on the MCU which is not the reset but right beside it, it needed to be on pin 14, at little solder bridge seems to fix it. I know the software has a --hwreset which is what toggles this and keeps from a total power down cycle. Anyway what I am noticing if I had an A2 unit that has boards like either one of these , anytime I stopped hashing for any reason I would have to reboot it, is that the case?
Thanks

Thank you very much for this, I have finally be able to determine why I have not been seeing the output that I have been looking for. Long story short the RPi that I have is the B model Revision 1, well turns out the terminators use Revivsion 2, there are 3 pins that hook to different spots on the RPi ARM processor GPIO 3,5 & 13, turns out the other CS signal I was looking for is on 13 and there is a reset on 5, on my RPi these are mapped to the camera interface connector. This also explains why it will not work on a B+ version even if you hooked the pins in the same place. All things aside I'll make up some pretty stuff on how to hook a blade directly to a RPi if anyone wants to test anything in a simple matter, shoot people may already know this info but I couldn't find anything out there.

Now just to find heatsinks and a great price, and decipher the controller board a little more, pretty much guessed about the LVC244s and the HC138 but wasn't expecting the LVC11, now to try and make a schematic of this board.

Well today's results were good, I got the board hashing directly connected. A couple things I have learned it does take much to break it, most of the times just stopping cgminer kills the comms to the MCU, at first I was turning everything off and back on then realized that the MCU reset it broke out to the 10 pin connector and just needs a low pulse but for some reason there is a resistor missing to actually connect it out to the world. I don't know what pin actually toggles this off the RPi yet.



Test setup. Don't have a heat sink so I attached it to a very massive piece of aluminum gets me a little bit of testing.


Hashing test


I don't know if anyone is really interested in any of this but it is kind of fun to tinker with I am wondering what the code on the STM32F103 is now... humm