Thanks for giving us the update.

Heatsinks on 17 series miners are not glued on, they are attached with solder. Most of the time I see heatsinks that have fallen off, it's because the copper plating on the top of the chip has delaminated. Without the plating, it is not possible to solder it back on. You could try to glue them on with any thermally conductive adhesive. Can't say if or how well it will work. I always replace chips that have lost their plating.

The thermally conductive epoxies I was able to find technical datasheets on had much worse performance than solder. Low temperature solder (Sn60Bi40) solder conducts heat about 30 times better than the epoxy. So the epoxy might not work well enough and the chip may overheat. If it isn't on one of the 4 chips that is monitored for temperature, it could get hot enough to permanently damage the board, or melt the solder so the chip will slide out of position.

Here's a link that has the thermal conductivity of many solder types: http://alasir.com/reference/solder_alloys/

And here's a thermally conductive epoxy: https://www.mgchemicals.com/downloads/tds/tds-8329tff-2parts.pdf

Nope, no need to program the eeprom unless using stock firmware.

It will detect and read temp sensors.

If the pic is bad, nothing will work, so there isn't really anything to detect.

As far as how to replace chips, there is no correct answer. But I no longer tin the chips with a stencil. I was never able to reliably get a chip on that way and had to drag reflow the connections with an iron, or add solder paste and clean up after. So I started just adding some solder to the main power pads and not bothering with any solder on the pins. I leave solder on the PCB, and then add flux and drag-reflow the pins with an iron after flowing with a hot-air tool.

Every set of tools and every person's skills are different though, so you really just need to try different approaches and figure out what works the best for you and the tools you have.

I have seen a few heatsinks come loose without taking the copper plating with it, but for the most part I'd say the main reason why the heatsinks come off is because the copper plating delaminated. The delamination is a manufacturing defect, you can clean solder off without taking the plating off if the chip is not defective. This is what chips with delaminated plating look like:



Sometimes the heatsink was never attached very well, and the solder will just break off, or the chip got hot enough that the solder melted. So in those cases, the solder will still be on the chip, and you can clean it up and re-attach the heatsink with no issue. I also don't generally remove all the solder on the chip, I just add a glob of flux and reflow the solder with a soldering iron (with a good-sized chisel tip) so it is smooth and covers the whole chip. If some or all of the copper plating is gone, then the solder will just bubble up and it will be impossible to get it to flow over the whole surface of the chip.

I don't think that is an accurate way to measure what fee they are charging. I think you'd need to compare the #s to the stock firmware running the same frequency/voltage to get an accurate comparison. Also not sure that 1 day is long enough to get a measurement accurate enough to see a 1% difference. Just look at the daily average of any miner over time. I just checked the daily average graph on a worker on Viabtc and it varied >3% over the past couple of weeks.

The purpose of the copper plating is to accept the solder, without it the solder won't flow. Think of water on a hydrophobic surface (like a windshield treated with rainx). So you could probably attach a heatsink with solder, but it would not make good contact because the solder will not flow over the whole surface of the chip. You could attach a heatsink using a thermally conductive adhesive, but that has other issues, like how to remove the heatsink if there is an issue with the chip or its connections.  So I always replace chips that have lost their copper plating.

I haven't done a lot of swapping chips between boards, most of the time I'm not going to take the risk of installing a chip that may be faulty. Just not worth the cost of a new chip to spend the time of putting it on only to have to remove it and reinstall another one. But the only advice I have for that is that you should either clean the solder off the chip or the board. If there are solder bumps on the chip and the board, it is very difficult to get the chip aligned, and when you heat it up it will shift. So I'd probably leave the solder on the chip and use solder wick to clean the solder off the board.

Workbench update...

I added a benchtop DMM to my workbench a few weeks ago ... I'm loving it so far. Don't really need the accuracy, but the time to make measurements is way less than the portable I was using and that time adds up.

Also pick up a Hakko FX951 soldering iron. It works great, but I'm now curious about a Thermaltronics iron (which is basically Metcal clone, started by previous Metcal employees).



Looks like AD and AG chips are all gone, but AI chips are still available, https://www.aliexpress.com/item/4000098330595.html?spm=a2g0s.9042311.0.0.d9754c4d7ET3MD

Use 3rd party firmware, they ignore the settings in the hashboard eeprom.

Did you mix and match hashboards from some different miners? I think the "Voltages are different" message means the voltage settings that are programmed into the eeprom on each hashboard at the factory are not the same. The same supply is used for all 3 boards, so they can't be set differently.

Your best option is to just use 3rd party firmware like thierry4wd or vnish. They let you set the voltage and frequency, so it will just ignore the settings programmed into the hashboard.

I think to get it working with factory firmware you'd need to have the eeproms on the hashboards reprogrammed so they match.

Replacing those caps is pretty easy. Only thing you need to be careful about is to make sure you flow the solder so a good connection is made. The pins are tied to big power planes so getting them hot enough can be a challenge.

They are just there to help filter out voltage variations caused by the quickly changing current usage of the hashboard. It might work fine without them. It would not damage anything to try.

With stock firmware you can sometimes mix and match hashboards, but not all hashboards will work with each other. I believe settings stored in memory on the hashboard need to be compatible. It won't damage anything to try, if it doesn't work the miner just won't get through it's initialization.

With aftermarket firmware, you can mix and match all you want.

It's possible it might still work, just depends on if the die was also damaged and not just the top package. Even if it works though, the thermal performance isn't going to be great. I've never bothered testing any asics that are broken like that, I just replace them.

You could try to test it before even attempting to put the heatsink back on. Just keep an eye on the kernel log while the miner is booting up, and shut it down if it actually gets past the "found 30 asic" message. If it says it found 0 asics, or less than 30 for that chain, then most likely the chip is not salvageable.

My experience repairing these so far has shown that the faulty miners typically have multiple issues, and all the issues don't always present themselves. I run miners several days after a repair to make sure they are stable, and nearly 100% of the time they fail again, normally in different ways, within a few days. So not surprised to see a lot of reports of "repaired" miners arriving broken.

About that, I recently called the new Bitmain repair center that opened March 1st about repairing a bunch of broken APW9s. They told me that they were only doing in-warranty repair at the moment because they already have more requests than they can handle. And HM Tech is only taking business from people with 25 or more dead miners. All these repair centers have a constant backlog of 17 series miners waiting to be put back online.

With most estimates of the 17 series failure rate between 10% and 20%, that's a pretty good chunk of hashrate that could be brought back without any need of TSMC.

The Russian one allows you to work much more efficiently. It allows you to hook up and start running a test in seconds, and then it will continue to exercise the board indefinitely so you can measure the test points on the hashboard without having to re-start the test over and over. The standard Bitmain tester like Zeus sells will attempt to get the asic count 3 times and then stop, requiring you to wait 30+ seconds for it to finish its cycle and before you can tell it to start again.

The Bitmain one does run a test pattern through the chips at full power to test out the actual hashing at speed, the Russian one does not do this. So if you want to detect low-hashing chips, you'd need to run the board in a miner with aftermarket firmware that will show the performance on a chip-by-chip basis (like vnish).

I use the Russian one >95% of the time, and just occasionally use the Bitmain one.

Also, the Bitmain style tester is just a S17+ control board with special firmware. The display screen and button on the test fixture they sell are not needed. The button is just hooked up to the reset button on the control board, and the screen doesn't show anything more than you get from the terminal output. So you can save yourself some $ by just buying a control board and getting the test firmware. You just need to get a serial to USB adapter like this https://www.sparkfun.com/products/15096 and hook it up to the serial signals on the control board.

One last thing, I think many people might think that the test fixture will tell you what chips are bad, and then all you need to do is replace the chips it identifies. That is not the case. Occasionally it might give you a clue as to where an issue is, but most of the time it will just tell you it found 0 chips. The main purpose for it is to exercise the board so you can go in and trace signals with a DMM or Oscilloscope to locate the issues.

If you do go for a stereo microscope, a nikon with a 0.5x auxiliary objective gets you ~180mm working distance. Mag with the 0.5X should be more than enough, and the extra working distance is key if you want to try to use a hot air tool. You can find some pretty decent deals on ebay for a nikon on a boom stand.

I have pretty good vision, but I still think it's overkill. I'd recommend spending the $20 on some cheap 3.5X magnifying glasses like I posted a link to before and give them a try first.

I don't have an official answer on the difference between AD, AG, AH, AI... but according to Zeus they are all interchangeable. I've used the AI successfully for S17, S17 Pro, T17, and S17+.

From the Zeus website:

I've also ordered successfully from Zuesbtc. It's a bit of a process though to get an order in through whatsapp.

I ordered 100 1397AI chips from IOTstore on aliexpress on Sunday and they have already shipped and are due to arrive on Friday.

Did you order from the store from my link, CN888 Store? I successfully ordered 5 times without issue from them. I've also ordered from here too.

Not sure a stereo microscope will be all that useful. I have a Bausch and Lomb on a boom stand, similar to this, but I don't use it for hashboards. Good for inspection after rework, but being fixed in a vertical position isn't ideal for placing/moving components when reflowing.

So far, what I've ended up using most is something similar to these: https://www.magnifier.com/headband-magnifier-eye-glass-style-5-lens-led-eg1li.htm

If I was going to drop >$500 on something, I'd go for some decent surgical loupes rather than a stereo microscope.

Just talked to someone at the Bitmain New Hampshire repair site. The guy there told me they are only doing in-warranty repairs at the moment because they have too big of a backlog.

Can't say I'm surprised. With the current high profitability and the poor reliability of the 17 series miners there aren't enough repair techs to service the demand.

You could try the new Bitmain repair center. I don't know if I'd call them reputable though with what some people have reported about them not returning miners.

Someone posted this in the marketplace a few days ago, no idea if/how legit it is. Looks like it may be a Russian company, but they have a location in Colorado.

https://service.myrig.com/

You might also want to look at ROI in terms of BTC instead of $, especially if you are a hodler. If you assume difficulty will increase, then the ROI will take longer in BTC.

If you assume that price and difficulty are going to double in a year and you'll ROI in $ in a year, the $ value of the BTC you spent (or didn't buy) would have been worth double while your miner just got back to 0.

Lots of people have issues with Nicehash, try a pool like kano, viabtc, f2pool, etc... first before trying anything else.

also, with nicehash you should add #xnsub to the end of the pool URL if you haven't already.

I don't see anything wrong in the log.

Does the log show any more messages after the hashrate drops?

When the hashrate drops, what does the miner status page look like?

Have you tried different pools?

You could also try flashing with an SD card to make sure you are getting a clean install.