Yes, the Asic Repair tester can burn the eeprom if you buy that option, but I don't think you can tell it what settings to write. So it just writes some default settings. If the hashboard doesn't work well at those settings then the stock firmware will work, but not very well. I haven't asked if there is a way to change the frequency and voltage setting that it writes, haven't needed it since the miners I've been repairing are pretty much all using vnish already.

I have mixed boards a few times with S17s running vnish and have not had any issues.

To run stock firmware, I believe both the Pic program for all the boards need to be identical and the tuning data stored in the eeprom on the hashboard needs to be compatible. The pickit does not program the eeprom, that is a chip separate from the pic. I believe the tool from Zeus writes the eeprom, not the pic. It lets you read the eeprom settings from one board, and write them to another.

My understanding is that the different pic program revisions require different communications api, and at least in the stock firmware, it only supports one api at a time. That same limitation might also apply to vnish.

With some solder paste and a hot-air tool of some sort applying a package like that pic is pretty easy. You can be pretty messy with the solder paste, and when you heat it up to the melting point with a hot air tool the solder will flow to the pins, so no need to worry about individual leads of the package. You can go back after reflowing the solder to clean up any bridges that might be caused by excess solder. Recently I've been doing that by applying flux over the pins, applying heat with the heat gun, and then running the blade of a x-acto knife over the solder, this does a great job of sucking up the excess. But many of these techniques are all personal preference, and what works for me might not work for you. So the best is to find some old electronics to rip apart and practice with. Find similar packages and remove and replace them. better to practice with worthless junk than a hashboard that could be worth a lot if/when repaired.

Here's an example of removing/replacing using an iron https://www.youtube.com/watch?v=HGgat5IVfFE

I normally do this by putting down a bead of solder paste over the pads, place the part, and then heat the whole thing up with my heat gun to reflow it all at once.

More like this: https://www.youtube.com/watch?v=f_yFDpSTfao

---

And the weller/pace/hakko/... stuff gets pretty expensive. Some is worth the $, but some not. I keep an eye out for a decent deal on used items on ebay.

But I'd just start with some lower cost ones, they may not have as high of a build quality but as long as you don't get complete no-name brands most will do the job just as well as the $$$ versions. For a hot-air tool, you just need to be able to set airflow and temperature and need something around 1000W. For an iron, you need one that you can set temperature. Those are the only necessary requirements I think. I have never used de-soldering tools, not sure they are of much use for surface mount components.

So for a decent quality low cost iron, I use this: https://www.tequipment.net/HakkoFX888D-23BY.html?v=118031

And as I said earlier, I just have been using a heatgun with temperature and airflow settings , but a reasonably priced hot-air tool like this one I'm sure would work fine

https://www.tequipment.net/Quick/861DW/Desoldering-Equipment/Rework-Stations/?v=7450

Here's a pretty in-depth review of it: https://www.youtube.com/watch?v=N_yHkrVYrBE

In fact, I just ordered one of those after watching that review.

If you have 5c power and expect the diff and price to be flat for the next 6 months, then that $10,250 miner will profit a bit more than $4000 in the 6 months you'd need to wait for the September batch. So looks like the only way you can rationalize spending the $ to get it now is if you're convinced the price is going to keep going straight up and the diff is going to be flat.

Did these ever come up for sale? The Bitmain webpage still shows only the sold-out August batch. I'm interested in seeing what price they are asking for a miner you won't see for 7 months. And do they require you to pay 100% upfront?


I think how they have done this in the past is at the hashboard level. Probably just too costly to try to characterize each chip individually. They test every hashboard individually at a series of different voltage/frequency settings and then combine "compatible" ones to get Pro/non-Pro, or different hashrates. For instance, the S17 and S17 pro hashboards are identical and use the exact same ASIC chip. For the S17+ hashboards I've seen they even have a label on the hashboard indicating what test level it passed at, L1 through L7. This is why aftermarket firmware can get such great results, they tune at the chip level instead of the board level.

There is always a range with these things due to process variation at the fab. Intel does the same with it's CPUs, the 2Ghz and 3Ghz models are exactly the same chip, one just hit the top of the curve and the other the bottom. So not really a quality issue, more just process drift and transistors that switch just slightly slower and/or less efficiently.

Even if you're running your S9s in normal mode at 90W/TH, and you're paying 10c/kWh for power, the price would need to fall below $37,000 at the current diff to be unprofitable. So I don't think anyone is turning off gear because it is not profitable anymore.

But I could see people taking gear offline to sell it. It would be pretty tempting to offload all your >=55W/th gear when people are paying >50$/th for it. So it could be that every time we have another pump in the price, another round of big mining operations take the opportunity to cash out and then the network hashrate falls while the gear is in transit to it's new owners.

The problem with the adhesives is that if just re-attaching those heatsinks doesn't work, it's gonna be a pain in the ass to get those heatsinks back off. So if all you're comfortable doing is re-attaching the heatsinks and if that doesn't work you're giving up, that's a fine route to take. But if you'd want to continue down the rabbit hole to try to repair it, I'd suggest re-attaching with solder.

https://bitcointalk.org/index.php?topic=5217949.msg56114781#msg56114781

I now mount all heatsinks the way I describe doing the S17+ in that post. It is just too easy to shift the chip when applying the heatsink if you heat the heatsink and place it on the chip rather than heating the heatsink up in place.

You'll need a test jig to have a reasonable chance at getting boards back up. The standard Bitmain type, like this, you can get at a bunch of different places. It is a S17 control board with special firmware that will exercise the hashboard so you can probe signals with a DMM or oscilloscope. They are pretty cumbersome and slow. I bought another tester from https://tester.asic.repair/en. It is more expensive and doesn't do as in-depth of a test as the normal Bitmain type, but it is 10X faster. With it you can connect the tester and run a test in less than 10 seconds. It takes a few minutes to do the same with the bitmain tester.

---

Here are some replacements for other parts:

Boost circuit controller https://www.digikey.com/en/products/detail/monolithic-power-systems-inc/MP1517DR-LF-Z/9433294

Boost circuit 1.8V regulator https://www.digikey.com/product-detail/en/monolithic-power-systems-inc/MP2019GN-Z/1589-1557-1-ND/9433323

1.8V regulator https://www.digikey.com/product-detail/en/on-semiconductor/NCP114ASN180T1G/NCP114ASN180T1GOSCT-ND/6560646

25mhz oscillator https://www.digikey.com/product-detail/en/diodes-incorporated/FK2500065Z/FK2500065ZDICT-ND/9952907

For ASICs, I've ordered from Zuesbtc and Aliexpress.

https://www.aliexpress.com/item/4000098353215.html?spm=a2g0s.9042311.0.0.11884c4dXGgZuU

https://www.zeusbtc.com/ASIC-Miner-Repair/Parts-Tools-Details.asp?ID=165

Don't know if I'd just plow ahead and start replacing things, the boost circuit they said failed has nothing directly to do with the pic so there's likely more wrong than just the pic. The one board I've repaired that had a blown pic also had every 1.8v regulator blown (12 of those on the board), and the boost circuit controller.

But the pic it isn't a very difficult part to replace. Just search for some video tutorials for surface mount rework.

This is the part: https://www.mouser.com/ProductDetail/579-33EP16GS202TE-SS/

Mouser also sells the pickit, but only the newest one. You can get the pickit 3 for cheaper elsewhere.

https://www.mouser.com/ProductDetail/579-PG164140/

This is the schematic of the connector where the 3.3V and i2c signals that the control board uses to talk to the pic.



Here's the schematic for the pic



I got both of those from the repair manuals you can get from zuesbtc here:

https://www.zeusbtc.com/News.asp?Sort=FILES+DOWNLOAD

You can run the guides through online translators to get a reasonably understandable translation. Some of those manuals also have info on using the pickit.

The error messages from chain 0 are from not getting any response from the pic microprocessor on the hashboard. This is a pretty basic function, the board only needs power supplied from the control board and a functional pic to get a response. Without the pic, the control board can't turn on the power to the ASIC chips, so there is no chance of getting anything working. Could be that the pic got fried and needs to be replaced.

All the S17 series (normal, pro, +, and E) have 3 hashboards and they will operate with fewer if you unplug the data cable for hashboards you want to disable. S19 should be the same although I haven't actually seen one yet.

For a plain S17, 1 hashboard does 16 to 19 TH at around 45W/TH, so 720W to 855W per hashboard.

Not likely to find any at reasonable prices though. And be warned that the 17 series had a lot of quality control issues so high failure rate.

I think it was certain batches, or maybe miners made at certain facilities. I have 7 S17 pros from Malaysia, and 2 out of 7 failed in less than a year. But I also have 10 S17s and 3 S17 pros from China, and I haven't had a single one go down.

Even if you do have the tools and skills to repair them, many times there are just too many issues to make it worth the effort. Many of these miners have flaky connections everywhere.. so you fix one issue and it runs for a few days until another issue springs up. Not worth the repair when it becomes a full time job to keep a single miner going...

Here are some plots for the last 3 epochs I made using the data from LoyceV's file. The green and orange are the probability of more than and less than the number of blocks at that point in the epoch assuming the actual hashrate for the whole epoch was flat. So when they diverge a lot, it suggests a better chance that the hashrate actually changed and is not just the error you'd assume.

---

And breaking out my old profitability table ... looks a lot different than the last time I posted this when S9s were "dead"... still dead at my 21c/kWh I pay at home though. But at least I can test out the S17s I'm repairing without losing $.

If someone was shutting down 6EH of S17s at 5c power, they'd only be giving up $69,700 per hour.... lol.

OK, I think I see what you are saying.

If you look at a the chances of being ahead or behind at each block of an epoch, you are more likely to be equal to or ahead than behind through the whole epoch, but getting closer and closer to 50/50 as you get to the end. But if you look at just the chance of just being ahead it changes the picture drastically.



Well, the data is available, so maybe if I get ambitious I'll look at the blocks from the last couple epochs to see what we might be able infer. All you need are the timestamps of the blocks to re-create the % ahead/behind.

Um, no. If the hashrate was 100% flat, over long enough time scale the average diff would be flat. By definition, the average time to find a block for the given hashrate is 600 seconds.

The Poisson distribution gives you the probability of X events happening in a given time period when you know on average Y events happen. So if on average we expect one block in 600 seconds, you can calculate the chance of getting 0, 1, 2, ..etc for the time period instead.

Although the chance of finding 1 or more blocks in 600 seconds is 63.2%, and the other 36.8% you find 0 blocks, that does not mean that the average expected time is lower. The chance of finding 1 or more is higher because the "or more" contains the chance of 2 through infinity number of blocks.

On average, it still settles to 1 block in 600 seconds.

This is the Poisson distribution that shows the chance of finding a number of blocks in 600 seconds:

blocks    chance
0      0.3678794412
1      0.3678794412
2      0.1839397206
3      0.0613132402
4      0.01532831005
5      0.00306566201
6      5.109436683E-4
7      7.299195261E-5
8      9.123994075E-6
9      1.01377712E-6
10      1.01377712E-7

The upper cumulative distribution gives the chance of finding X or more blocks in 600 seconds

0      1
1      0.6321205588
2      0.2642411177
3      0.08030139707
4      0.01898815688
5      0.003659846827
6      5.941848176E-4
7      8.324114929E-5
8      1.024919667E-5
9      1.125202598E-6
10      1.114254783E-7

And the lower cumulative distribution gives the chance of finding X or fewer

0   0.3678794412
1   0.7357588823
2   0.9196986029
3   0.9810118431
4   0.9963401532
5   0.9994058152
6   0.9999167589
7   0.9999897508
8   0.9999988748
9   0.9999998886
10   0.99999999

The other way to use Poisson is based on rate of success,  if your average expected success rate is 50% for finding one block in a time period (like finding one or more blocks with 50% diff work, or in 1/2 the expected time), the probability is 39.3%.

Finding one or more blocks in 1/4 of the time would be 22.1%, and finding one or more blocks in 2X the time would be 85.4%.

FYI, here are a couple of online Poisson distribution calculators,

https://keisan.casio.com/exec/system/1180573180
https://stattrek.com/online-calculator/poisson.aspx

Excel also has a poisson function, POISSON.DIST()

Back to my earlier post about the chances of being 10% ahead after 700 blocks when the hashrate was actually flat, that is the chance of getting 700 blocks or more when on average we would expect 636 (not 770 vs 700 like I said before), that would be 0.649% (1 in 154 times). So you might expect that to happen once every 6 or so years... doing that 3 times consecutively is 1 in 3.6 million.... so once in 140,000 years... lol.

Well ... if I'm doing/applying my poisson calcs correctly, getting 770 blocks or more when the expected # for the actual hashrate is 700 would only happen 1 in 209 times. Doing that 3 times consecutively is not likely unless there's something going on with the hashrate.

10% ahead after the first day is more like a 1 in 8 chance.

If the PSU fans are blowing in the opposite direction then they are taking in the heated air from the miner exhaust, and heated air from the PSU exhaust is getting sucked into the miner intake.... so not a good scenario. Pretty sure that isn't how it's supposed to work.

1.5 years later and still too lazy to put it on ebay, so still available. Prices are up... so now back to $250 + shipping + escrow.

Maybe one of these would work...

https://www.newegg.com/noctua-nf-a14-ippc-3000-pwm-case-fan/p/N82E16835608048?Description=fan&cm_re=fan-_-35-608-048-_-Product

Probably lower speed than stock, but if you're going to try to run lower power to keep the noise down maybe it would work.

Other options seem a bit too expensive:

https://www.mouser.com/Thermal-Management/Fans-Blowers/_/N-axg88?P=1yx2dyrZ1z0sdx2Z1z0xjmu&Ns=Pricing|0

Yeah, your 12kW of continuously running miners is probably using up > 1/2 the capacity for your street....  Just because you have 200A service doesn't mean you can run 200A continuously. The power company sizes the transformers based on average usage, so if everyone ramped up to using their full 200A service at the same time the transformer would go down and take the power for half your street down with it....

If your interested, you can probably trace the power lines coming out of your house and find the transformer hanging on the pole that is powering you and your neighbors. They are normally labeled with the KVA capacity.

OK, the chip in the bottom of the photo is physically damaged, the heatsink was hit with enough force to break the top asic package off. Pieces of it are probably still attached to the heatsink. Not going to be able to glue that back on and get it working. The chip may be functional electrically, but I don't think you'd be able to get a good enough connection to the heatsink to allow it to operate without overheating. That chip will have to be replaced to get the board working.

The other two are not physically damaged, but the copper plating that interfaces between the asic package and the heatsink has delaminated. Normally, that copper will accept the solder they use to attach the heatsink. Without it, the solder won't flow and will ball up creating a bad connection that won't transfer the heat well enough. Gluing those two heatsinks back on with a thermally conductive adhesive could work though.