T17/S17 malfunction: cases, solutions, remedies, RMA history

[wndsnb]

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.

[1714950367]

1714950367

[1714950367]

1714950367

[1714950367]

1714950367

[1714950367]

1714950367

[favebook]

You might be right, I could be very wrong. If you are right, then that would mean that there is a different problem and my initial thought was totally wrong. PSU might still have caused initial droppage of chips and I am nowhere near closer to narrowing down what is bad. I will just RMA it and let Bitmain decide what they did wrong.

[wndsnb]

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.

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.

[favebook]

After careful and thorough examination of hashboards, I have found in addition to oxidised copper plate (thank you wndsnb for clarifying that), there was a heatsink which was misaligned in comparison to others see pictures below:



Note that they were touching at top most part but with a gentle push of finger I made space between them in order to test if the hashboard will power on, but it was futile attempt. The bottom part of loose heatsink is around 0.2mm away from nearby one, but as top part was touching, there was probably a shortage there.

Moving on, after the futile attempt at "fixing" it, I gently pushed the heatsink in case with my middle finger and this was the result:



May that hashboard rest in peace.

Now, I was thinking about 2 other hashboards that were working fine during the hottest day of summer (35C+ ambient) but as soon as ambient temperature dropped below 5C (few weeks ago) they started reading 0 out of 65 chips or 12 out of 65 chips (it warried from time to time). I figured maybe it had something to do with Thermodynamics law of expansion and contraction in hot and cold time respectively. So I took my Z11 (which is underclocked for better efficiency and therefore doesn't produce that much heat anyway) and placed it in front of my S17+ intake (picture below). And I was astonished by result. After first boot, both hashboards found all 65 out of 65 chips and started hashing away instantly.



To further secure my unit to hash as much as possible and as long as possible, I updated to newest firmware which seems to fix the exact problem I have but I was unaware of this before I tested the method I noted above.

1. Enhance the working stability of the miner under special conditions (e.g. a certain range of low or high ambient temperature environment).

I hope this helps someone in future.

[wndsnb]

Moving on, after the futile attempt at "fixing" it, I gently pushed the heatsink in case with my middle finger and this was the result:

https://i.imgur.com/8Nayhat.jpg

May that hashboard rest in peace.

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.

[favebook]

I have never done soldering before and the only experience with heat gun that I have was bending hard-line tubbing. I wouldn't dare to do that yet, I might buy broken S9s to learn how to do it on them but I would rather RMA this unit right now.

I could provide you a better and zoomed in pictures of chip and heatsink.

Solder snapped with heatsink for sure, the only thing that I could see on bottom of the heatsink was name (or part of it) of the chip, which might or might not indicate that it took a part of chip as well.

That chip group did have a problem with overheating now that you mention it, but never above 90C, it was usually in mid 80s and during winter in low 70s.

It is one of the last chips on the board if I remember right.

During my initial testing when I didn't even open the unit, that hashboard would at least pop up in voltage check. But now after loose heatsink it doesn't even show up in Kernel Log which is normal but it might mean that it is possible to save it, but as I said, I am not up to that challenge.

[NotFuzzyWarm]

So instead of soldering, use a good epoxy thermal adhesive to reattach the heat sink... Assuming the chip itself was not damaged, should work just fine.

[favebook]

So instead of soldering, use a good epoxy thermal adhesive to reattach the heat sink... Assuming the chip itself was not damaged, should work just fine.

Would you recommend any specific one? Preferable purchasable in Europe?

[NotFuzzyWarm]

Do a search for Arctic Silver or Arctic Alumina on your fav site. Key point is that it be a thermal epoxy so it holds the heat sink in place and is NOT just a thermal paste. Thermal pastes and stick-on thermal pads will not rigidly bond so things can move.

[favebook]

Sadly there are no Arctic Silver or Alumina in my country. There are only epoxy resins in my country and they are used for something totally different. Looks like I will have to pull cords to get it from Germany/Italy or I will have to order from Amazon.

Thank you for your help tho. I will try it before RMA-ing the unit.

[mikeywith]

Would you recommend any specific one? Preferable purchasable in Europe?

Read my topic, I had gone through a similar problem, I ended up ordering Arctic silver form this link https://www.amazon.com/dp/B0087X7262 for $17, now the price has gone up to $37 which is way too much, actually even $17 is a lot for 7g, if shipping from China is an option and doesn't take forever, then get the black glue from a place like ZeusBtc https://www.zeusbtc.com/RepairDetails.asp?ID=86, you can actually find it cheaper on Aliexpress (not sure if it's the same quality).

If you end up ordering the "black glue" make sure whatever forwarder you use will be able to get it to you within 10 days, the guys at Zeusbtc told me that it can't be stored out of the fridge for more than 10 days.

[wndsnb]

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

[NotFuzzyWarm]

If using thermal adhesive, be aware that most 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.

^^ Not necessarily. The Arctic Silver brand specifically state they are electrically non-conductive though do have a small capacitance value. Just make sure whatever you get states that it is electrically non-conductive.

When using any thermal compound that is conductive one trick often done is to simply apply a thin line of RTV (silicone glue for electronics) over the IC pins to keep the thermal epoxy from touching them.

[wndsnb]

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.

[Pendrak]

What thermal paste they use for attach those heatsink?

[wndsnb]

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

[NotFuzzyWarm]

As wndsnb said it is NOT thermal paste. Thermal paste does not mechanically bond things (heat sinks) to chips - it only fills gaps/voids between surfaces and relies on clips or other means to hold things together. Same applies to self-sticking thermal pads. Either solder or a thermal epoxy adhesive MUST be used or the heat sinks will eventually fall off.

[mikeywith]

Thermal cooling adhesive like the one from Arctic will do the job as well, the cheapest would be the "black glue" from China, I am not sure if that's more of solder or thermal adhesive, but it works, and it's dirt cheap, I tried to look for an alternative in the local market but with no luck, I am not even sure how do people here go about fixing heatsinks of other electronic equipment or maybe I didn't know what "material" I should be asking for since the words "solder/adhesive" didn't ring the bell, they would show me thermal paste used for CPUs and be like "this should work".  

[Vezza]

Code:

nand: WARNING: pl35x-nand: the ECC used on your system is too weak compared to the one required by the NAND chip

Does anyone know why I get this thing up?

[Breeze]

Hi guys
Which hashboard is hashboard 1 (chain 0)?
Closes to the PSU for example.
Maybe @mikeywith can help me?