Bitmain's Released Antminer S9, World's First 16nm Miner Ready to Order

[allinvain]

I have noticed that the temperature sensor on S9 boards has an unusually high failure rate and it's taking down entire hashboards with it. When the sensor goes bad the board no longer hashes. This has happened on numerous occassions with our customers' gears. Does anyone know by any chance if there is a way to fix this. Does someone know where this temp sensor is located (picture would be really helpful) and if it can be swapped out/fixed/repaired, etc?

The temp sensors are actually built into the chip itself. Problem is they only pull data from chip 63 so if that one has a problem you are out of luck. It will be fixed under warranty but there is basically nothing that you can do as an end user to fix it.

Oh crap, that's the worst case scenario. Dammit Bitmain why you go so cheap on the design/build quality? Do they not realize that if the miner dies out of warranty they are just basically screwing their customers out of ROI. Should these things not be built well enough/smart enough that they will last more than 6 months before killing themselves.

And how many of those (percentage-wise) are in a properly ventilated area that's always under 77F?  

Good point. But sometimes it's not the heat that kills them cause we've had board die in the middle of the North Canada winter. All it takes is one or two traces to become undone due to poor sodder or lack of sufficient sodder and bam, chip is offline and bam temp sensor is offline and bam blade is entirely offline. Also according to bitmain specs these are supposed to be able to handle fairly high intake temps - ie a traditional data center environment. I think the reality is a pure and simple ugly truth which is that Bitmain obviously cut corners on design and build quality to wide their profit margins as well as make the miners ore affordable of us miners. I would honestly pay a bit more for a rock solid design implementation and especially one that is user serviceable to a larger degree.

-update- has anyone attempted to remove the chips off of an S9 board? I'm wondering what kind of specialized equipment we'd need.

[1713527423]

1713527423

[1713527423]

1713527423

[1713527423]

1713527423

[1713527423]

1713527423

[1713527423]

1713527423

[1713527423]

1713527423

[allinvain]

@Bitmain, would you guys be willing to offer an out of warranty repair service for this dead temp sensor chip issue? Even if it costs us $50 it would still be better than a dead ~$300 blade.

[fanatic26]

-update- has anyone attempted to remove the chips off of an S9 board? I'm wondering what kind of specialized equipment we'd need.

I have removed and replaced chips on an S9 board. You can remove them using long tweezers and a heat gun. The trick is to heat up the larger top heatsinks and use the twwzers to roll the heatsink back and forth, dont just yank on it. Then remove the bottom heatsink in the same manner. If you do this backwards and heat up the smaller heatsink first, when you go to remove the larger heatsink you will yank the chip right off the board. You can heat up the chip a bit and use a razor to scrape off the black goop. Then simply heat up the chip ( around 400-450 f on the heat gun ), pinch it from the sides with the tweezers and slightly wiggle until it pops off, keeping heat applied the whole time. Be very careful with the heat and tweezers as there are many small components you can heat up and knock out of place without even realizing it.

[fanatic26]

@Bitmain, would you guys be willing to offer an out of warranty repair service for this dead temp sensor chip issue? Even if it costs us $50 it would still be better than a dead ~$300 blade.

You would still pay out the ass for shipping + repair. Its the same thing as repairing any other out of warranty board, they arent gonna give you a discount just because you identified the issue for them.

[NotFuzzyWarm]

@Bitmain, would you guys be willing to offer an out of warranty repair service for this dead temp sensor chip issue? Even if it costs us $50 it would still be better than a dead ~$300 blade.

If in North America talk to BitmainWarranty in Colorado about it. Also have several locations in other parts of the world as well. https://bitmainwarranty.com

From a power electronics design view, having a temp sensor on the die is the best and most accurate thing to do. That said, with so many chips in these I wonder if there is way to just have the miner look at a different chip? Heck, all chips talk over a single serial bus so should just be a matter of reassigning which to take data from. Problem solved.

As for removing the chips... Check the s7 threads and things dealing with sidehacks 2Pac stick and upcoming Pod miners.

Assuming Bitmain did the same thing with the s9/T9 (looks like it) then fuggitaboutit. If ya know which chip to attack, maybe but it is a risky process worsened but another design point: They made damn sure it would be very hard for the heat sinks to get knocked off buy using a very tough thermal epoxy to bond the topside heatsink to the chip.

A heat gun to applied to the heatsink will soften it enough to pull it off. So far so good. Then you see that when assembled the epoxy has flowed around the outside of the chip itself and so far the normal solvents will not touch it.

Design plus for mechanics and increasing thermal contact area a bit. Con is near impossible to replace a chip without risking damage to the board.

edit: Just saw Fanatic's post above ^^ Bet reflowing the new chip on is fun with the fine lead pitch...
Cool but still would rather sent to CO. $14.95 to ship them a board and that is with it insured for $450 in case it takes a wander.

[NotFuzzyWarm]

re: removing the epoxy
Your mention of scraping got me thinking. Not heat - use canned cold-spray or LN2 if ya got it. Make brittle and should chip off nicely. Epoxies made to take heat usually do not like deep cold and embrittle easily, comes with the turf of what is intended for. Hmm....

[ComputerGenie]

Good point. But sometimes it's not the heat that kills them cause we've had board die in the middle of the North Canada winter...

I'm not saying that all failures are due to heat, just implying that, imo, the larger portion are operator controllable issues (i.e., overheating, dirt, etc.) and not the "fault" of Bitmain.

... Also according to bitmain specs these are supposed to be able to handle fairly high intake temps - ie a traditional data center environment...

With most manufactures suggested temps of 65-75F, I can't think of anyone that runs an efficient setup that's over 81F (even Dell, who has the hottest "highest temps", never suggest higher than "lower 80s"). 

[allinvain]

Thank you for the info guys, I very much appreciate it! 

[VentMine]

Yup mine came with Autotune...
But I checked my kernel just to follow along, mine does have "Check chain[5] PIC fw version=0x02" so you're saying I can go backwards to older firmware?

No, if it had "Check chain[5] PIC fw version=0x02" and it came from the factory with manual, then you could go back to manual.
If it had "Check chain[5] PIC fw version=0x03" and it came from the factory with manual, then you likely could not go back to manual.
If it came from the factory with auto, then you can only usefully go between autos. Some autos will run manual fw, but you'd likely be overclocking 50+% of the chips and risking burnout. Those that came from the factory with auto have a per chip setting (manual sets every chip with the same frequency) so that weaker chips can run on the same board as higher chips; one chip might be set @ 440 and others set at 675+, so if you went to manual and set anything over 440, then you'd be pushing the chip past it's known limit.

Thanks for the detailed reply, love learning about this stuff. I almost feel like going back to school, mechanical engineering now seems so boring.

[VentMine]

oh, re:

Typically when I want to reset the miner I just unplug it and replug it back in, and it normally hashes fine.

As in the AC power plug? Don't do that... As you may have noticed there can be a nice fat spark from the high-side DC caps charging up. Rather hard on the plug and socket pins.

If is a server supply there will be a DC-ON jumper or switch. On the Bitmain APW supplies it is the green jumper. Use that to switch the DC on and off.

OK thanks for this info.. oopsies 

[VentMine]

..Just spamming here lately I am..

So I got another question 

I am thinking about building an S9 farm in a shipping container. Yes, I know its been done before, its dumb, risky, etc etc.. When the container is shut down temporarily in the winter and the damn thing goes to -20 deg C or even -30 deg C (gets this cold in Canada), will the S9's likely be damaged? I mean, clearly thermal cycling is a problem due to stresses causing things to come loose, but will this temperature alone do harm? My plan would be to not start up the S9's until I have first warmed them up again to 0+ deg C.

[NotFuzzyWarm]

will the S9's likely be damaged?

YES!
For one you risk the large caps being damaged by the freezing. And then there are thermal stresses on soldered connections... Not so much 'making them come loose', more like it causing the solder to start crystallizing. Lead-free solder has enough of that issue at normal temps and deep cold makes it worse.

As a side note: Eliminating solder problems such as crystallization, whiskers/dendrite growth, and a few other things is why alloys containing lead were used in the first place. To-date, NOTHING has been found to take the place of having lead in the mix. Lead-free solder alloys have gotten better but will never be as good when it comes to reliability. Just the nature of the beast.
 
Install heaters to keep above 0C, preferably keep above 10C and you will be fine. Of course that means having vents with louvers that can be closed to keep the heat in when needed.

Before starting I highly suggest taking them up to around 15-20C for at least 15min or more before firing them up.

[VentMine]

OK thanks for that. Yeah, I'll have to install aback-up heaters, no way around it.

[HerbPean]

I have noticed that the temperature sensor on S9 boards has an unusually high failure rate and it's taking down entire hashboards with it. When the sensor goes bad the board no longer hashes. This has happened on numerous occassions with our customers' gears. Does anyone know by any chance if there is a way to fix this. Does someone know where this temp sensor is located (picture would be really helpful) and if it can be swapped out/fixed/repaired, etc?

The temp sensors are actually built into the chip itself. Problem is they only pull data from chip 63 so if that one has a problem you are out of luck. It will be fixed under warranty but there is basically nothing that you can do as an end user to fix it.

Oh crap, that's the worst case scenario. Dammit Bitmain why you go so cheap on the design/build quality? Do they not realize that if the miner dies out of warranty they are just basically screwing their customers out of ROI. Should these things not be built well enough/smart enough that they will last more than 6 months before killing themselves.

If there code were public, you could change it and pull the temp from any chip you want.

[Rodawg]

How come the BTC Rate on the bitmain site is constantly $100 below market?   are they keying of BTC-e?   

https://www.okcoin.com/api/v1/ticker.do?symbol=btc_usd

Code:

{
	"date": "1493413852",
	"ticker": {
		"buy": "1232.32",
		"high": "1322.1",
		"last": "1233.19",
		"low": "1209.0",
		"sell": "1233.53",
		"vol": "4889.364"
	}
}

So ridiculous--currently, sale price of S9 says "1089 USD ( 0.858 BTC )".  That is an exchange rate of $1269.  Even okcoin.com says, "Last $1,377.26 High $1,380.00 Low $1,338.24".  Conservative exchanges rates elsewhere (i.e. COINBASE.COM) show $1460.
 Therefore, BITMAIN Rate = Rate Elsewhere + $100 + $100 => BITMAIN is charging $200 more per miner or almost 20% premium!!

 

[Rodawg]

So ridiculous--currently, sale price of S9 says "1089 USD ( 0.858 BTC )".  That is an exchange rate of $1269.  Even okcoin.com says, "Last $1,377.26 High $1,380.00 Low $1,338.24".  Conservative exchanges rates elsewhere (i.e. COINBASE.COM) show $1460.
 Therefore, BITMAIN Rate = Rate Elsewhere + $100 + $100 => BITMAIN is charging $200 more per miner or almost 20% premium!!

 
[/quote]

On this topic--Has anyone paid BITMAIN via Wire Transfer?

That would be one way to get around this "$200 BTC Premium" they are charging...

Thanks in advance for the help.

[NotFuzzyWarm]

Yes people have used wire transfer. Um, it is NOT free. The banks usually on your end and Bitmain's end will both charge fees. Never used wire so don't know how much but someone here HAS mentioned/talked about it. Throw in the hassle factor/your time for going to the bank and I'd bet it works out the same either way.

edit: Actually, just ask your bank what their wire fee is...

[Minor Miner]

Yes people have used wire transfer. Um, it is NOT free. The banks usually on your end and Bitmain's end will both charge fees. Never used wire so don't know how much but someone here HAS mentioned/talked about it. Throw in the hassle factor/your time for going to the bank and I'd bet it works out the same either way.

edit: Actually, just ask your bank what their wire fee is...

$25 to $45.

[oawan]

It looks like the S9's currently for sale are from a previous batch, batch 12.  Does anyone have them?  How is the reliability?

[Sierra8561]

It looks like the S9's currently for sale are from a previous batch, batch 12.  Does anyone have them?  How is the reliability?

I have a few of the batch 12's no problems at all.