Can you post the log from the test fixture run showing the bad temp sensors?

If the issue is only a poorly connected heat sink, then it would only fail when the chip overheats. Going to be hard to find that.

You might try feeling the temperature of the individual heat sinks right after it fails to see if a heatsink is warmer or cooler than the rest.

This is a good place to start: https://www.zeusbtc.com/NewsDetails.asp?ID=182

There is a download link on the page for the full repair manual, but it is in Chinese. You can load it into google translate to get a somewhat understandable translation.

If you want to try to fix them yourself, then you'll need a test fixture, a multimeter,  adjustable heat gun, soldiering iron, ...etc. An oscilloscope is also helpful. I wouldn't recommend it unless you already have a background in electronics and have some experience doing surface mount rework.

They are attached with solder. This is what I am using: https://www.chipquik.com/store/product_info.php?products_id=470007

I'm finding that the hashboard I'm troubleshooting that had 5 heatsinks shifted and shorting adjacent rows also has shifted chips. And on top of that, it looks like too much solder was used when the heatsinks were originally attached and some of it has migrated down and is now shorting pins on some of the chips. Looking like I'm going to have to take off more than half the heatsinks before I can get this one back up.



Here's a thermal image of the board where you can see some of shifted chips in the upper left corner. Not really necessary, but a cool toy to play with. That is where all the shifted heat sinks were as well. This was taken when the test fixture was just querying the chips status to get an asic count, so no hashing yet. The photo is cool, but in this case, it was also easy enough to find those by just feeling the board.

They are using PPLNS with N = 5 diff rounds, so on average you'd get 5 blocks during a ramp up or down. With their current hashrate (~10000PH) that comes out to around a 10 hour ramp up, 10 hour ramp down. So in a daily cycle running only 12 hours, you'd ramp up in 10 hours, run 2 hours fully ramped up, then have a 10 hour ramp down. In that cycle, the pool would find 10 blocks on average. Granted, the blocks the pool finds near or during full ramp are going to pay more, but there are enough of them over the course of a few weeks I think the variance would be pretty low.

You'll see some variance, but I'd hardly call it gambling.

If using thermal adhesive, be aware that most some of them are electrically conductive. Be careful not to let any drip over the side of the chip which may short out circuits on the board.  

edit: just looked up arctic silver and it looks like that one does not conduct.

edit #2: ... looks like NotFuzzyWarm beat me to it

That may not actually be as bad as it might look. If the solder snapped without taking part of the chip with it, then re-attaching the heatsink with a heat gun might fix it (probably very slim chance though).

I think all of the shiny solder on the chip in your photo was not in contact with the heatsink, the part that snapped when you pushed on it would be the small dull section in the top left of the chip. Since only a small portion of the heatsink was actually making contact, it would have been overheating, so this would have probably caused issues and lines up with other suggestions that poorly attached heatsinks are a major problem with these miners.

So maybe, if the chip didn't already overheat to the point of failure, just re-attaching the heatsink might do it.

Well, the problem with this board was not the pic microcontroller. Turns out I was just not looking for its enable signal at the right time. The test jig I have only runs the board for about 1.5 seconds when it doesn't detect any ASICs, and I was measuring the pic outputs after the board had already powered down.

I was able to connect to the pic with the pickit4 and read the pic firmware out of a known good board and use it to verify the firmware in the bad board. So it does not look like there is a need to buy a download... a little sketchy that Zuesbtc is trying to sell it when you can read it yourself with the tools needed to program the chip. I haven't actually programmed a chip with the firmware I read out, so maybe there is some issue with doing that that I'm not seeing. Or maybe some boards do have the read protection bit set and I got lucky.

Looks to me like just oxidation or staining, probably happened during the reflow during assembly or in a cleaning process after.

I think that copper plate is connected to the positive rail of the PSU input, it is there just to increase the current capacity of the connection between the + input and the FETs used to switch the main power to the asics on the board. The FETs are the 4 8-pin chips just below the copper piece (Q1, Q2, Q5, and Q6 in the schematic mikeywith posted). Those are pretty beefy FETs, but they would have gone up in smoke long before a bare copper plate would have heated up to the point of discoloration.

If you have another PSU to try, then that's a good/easy thing to try, but I don't think the test you did indicates a PSU problem.

If I'm understanding what you did correctly, you originally had board 1 connected to chain 1, then you swapped the control cables so that board one is connected to chain 2. The problem stayed with the board, suggesting that the issue is in that board. Although it could just be that that board just happens to be more sensitive to a PSU problem than the others.

Maybe try running one hashboard at a time by disconnecting the control cables for the other boards? I'm guessing the same board will fail.

If it were me, I'd probably go with viabtc pplns since it has a 2% fee and their pool hashrate is large enough that variance is pretty low. Daily payouts may vary but by the time you get to a week or a month there isn't going to be much variance. Since fees are lower you'll end up making more profit in the long run.

I don't think those are chips, I think it is literally just a piece of copper. Possibly used to increase the current capacity of the signal. Can't really tell from your photo if it is just oxidation or scorch marks.

S9s are a tough one at current difficulty and price... most likely better off tossing the broken ones and buying used replacements. I think you can get them for under $50 at this point. But if you want to start repairing other miners practicing on S9s is a good way to start.

At a minimum, you'll need a multimeter, adjustable heat gun, soldering iron, solder wick, solder wire, solder paste, tinning stencil. A test jig is very useful.

Best place to start is probably just doing an internet search for miner repair videos.

Also, let us know what the repair places tell you. Interested to know what they charge.

Heatsinks on these miners are not attached using adhesive. Instead, the heatsinks are soldered directly to the top of the chips with low-temperature melting point solder (138 deg C). Using thermal adhesive to attach the heat sink to the chip may work, but it might be extremely difficult or impossible to remove it once it's on.

Zeusbtc has listings for independent repair centers on their website you could try contacting (bottom of this page https://www.zeusbtc.com/Repair.asp).

No, the psu requires 200 to 240 ac input and outputs between 12 and 15 volts dc. It will not run on 120v.

Started work on the 1st S17 from my host. 2 hashboards showing 0 chips. Turns out both of them have heatsinks that are misaligned enough to short to the adjacent row.



After moving this heatsink back into position, this board comes up and identifies all chips, but the test fixture is showing a couple of bad chips so it looks like I'll need to replace a few.

The 2nd board is still not coming up after moving heatsinks back into position. Looks like an issue with the pic microcontroller on the board not enabling power so I ordered a pic ISD (pickit 4) to investigate that further. Hoping that Bitmain did not set the read protection on the pic flash so I can read out the firmware to be able to program a replacement chip if necessary, but I'm thinking chances of that are close to 0. If not, it looks like zeusbtc sells a download for pic firmware for all Bitmain's miners.

Also, just another thing I noticed with both these hashboards was that there were insect wings (looked like moth wings) blocking a few of the input side of the heatsinks. Could be what caused some chips to overheat enough for the heatsinks to slip. The solder holding the heatsinks on is low-temperature solder, so it will melt at under 150deg C.

OP said 5 euro per MWh, that's 0.005 euros per kilowatt-hour...

Depends a lot on how much you can get hashboards for. Just took a quick look on ebay and the only S17 hashboards I see are $390 each and come from China. Better off trying to find a used working S17. Kind of surprised there aren't more dead or partially working T17/S17s on ebay with all the reports of QC issues. Maybe they are all waiting their turn at Bitmain repair centers... I sent my S17 pro to the California center in May and I still don't have it back, although I got a notification that a replacement is on its way a few weeks ago.

sorry, missed the post with the MWh.

... OK, so 0.13 Euro per kWh guaranteed vs 0.05 euro/kwh with a 55W/th miner (T17), 0.08 euro/kwh with a s17 pro, or 0.10 euro/kwh with a s19pro.

So even if you spend a ridiculous amount of money for the latest s19pro, you are still not making as much as selling to the grid.

I'm not going to do the calculations for LTC, but I doubt they are any better.


However...

If you can manage to use all the 800kW from your generator, the additional power costs you 0.005 euro/kWh!?! If the contract allows you to buy that power at that rate for mining crypto, then you have an opportunity if you can get enough capital to use all the 800kW + a ton more at 0.005 euro/kWh. I'd look for some investors to build a 10+MW facility ...

Yes, I understand that. But what is the price you pay, and the price you get paid in euros/kWh?