Hrm. Conformal coating does keep out the crud but does keep in the heat; I can see a 20c difference between having a thin layer of conformal coating on the water block to Neptune chip interface, sanding it off really dropped the temps.

Honestly though I don't think environment is that massive a deal with these kinds of units. Unless you're really running it in a dirt farm, or got the thing submerged in crummy water it shouldn't be that terrible. I had some thoughts on pictures of another company's miner that was refused RMA due to "environment damage" and thought "that's not unusual". Besides if it's a really corrosive atmosphere the fan will probably seize first, which will overtemp the device.*

A properly conformal coated board though can take a lot of environmental damage: I once fixed an electric car controller that had a leak from the water plate inside the controller box. Literally covered everything with oily, antifreeze laden gunk. Cleaned the boards, cleared out the fault maker (the pre-charge relay) and the board fired right up with no problems.

However in that case you have 400 volt DC lines on the board, and that kind of voltage will bridge gaps caused by gunk. 12 volts? Not so much.... .6 volts? Really not so much.

What do I use for home electronic repairs? TechSpral Fine-L-Kote SR silicon conformal coating spray. Light coat is all you really need.

Hope this helps.

* Then again I saw discussions about another that had been rne in in a closed room in a 110+ degree day in the Australian outback and they were complaining that it was throttling. No AC no nothing. Um.... that's what is called "way outside parameters". Too many manufacturers build for a 68 degree data center, which is not what the typical home has (there is a big difference between how much heat a radiator can dissipate between 70f and 80f)

Yes, that is it. It's actually a pretty busy little bee all things considered.


They are very similar boards, so what works on one should work on the other. Just sanded down and re-mounted the water cooler block, it's holding a nice 44c, which is actually 56c as read from the bottom of the board. Top temp on the bottom of the board is 68c under some of the DC-DC supplies, they do dump a lot of heat into the board planes.....

Thanks! Hm. This one is running at 50c, which is actually quite warm. Maybe ones that blow up run hot, I'll have to wait till I get another one.

In the meantime I now know what blows up Neptunes and probably titans (Note: NEVER move a hot or running unit, let them cool first) and can fix blown Neptune boards that don't talk back to the controller. It's a bit of a bitch to do, but for a blown Titan it's probably well worth it.

Now back to working on the water cooled Neptune. Having conformal coating on my water blocks (used to mount them on backs of boards) is cutting the efficiency. Likewise the chips run hotter than the temp sensors say, watch out....

Happy MacArthur Day! Had some free time so I spent it melting off one of the chips on a board that was a complete failure.

By "complete failure" I mean the board failed every attempt to talk to a controller. Shorts out the controller board basically, different values on pin 1 of the 10 pin adapter. Dead short in one of the DC-DC converters, black burn spot on the back.

Pulling the power supplies and caps did not clear the short so I pulled the chip. Note that these chips are *stupid big*. As in 380 degree bottom heat for 30 minutes. As in 470c top heat all over the chip before it finally came loose. As in lift off chip and have it fall back on the board because my picker doesn't have enough suction.

Oh well.

However I lifted the chip to preserve the top right side (the shorted one) and I can see the problem: The chip underside got so hot at some point it reflowed the solder. Resulting in the chip literally shorting itself. In this case I'm guessing one of those shorts was to the SPI control line which would render the board fucked.

Interesting. So much heat it literally melted the solder under it. No wonder why the board was charred...

Fixing this is going to be a serious bitch on wheels. Hm.

Just bought another game on steambitshop.com. The crazy thing is how incredibly good this site is. It's really a very nice way to get steam games without dealing with steam points and stuff.

Well, I bought a Steam game for 1.5 bitcents. I know, sometime when that would buy a mansion I will question my decision, but it was the quickest option at the time....

Quick question for anyone listening: Can you touch with your finger the FPGA chip on a running Neptune/Titan controller board and tell me if it's cold or warm? If warm, how warm?

Thanks and happy holidays!

why an inverter? Almost every miner known to God runs on 12 volts. Just bus that straight to the battery (use a 12 volt 30amps contactor tied to key_12 volts if you want it off when car is off) and go to town.

Well it may be harder to overvolt Titans. Scrypt has all sorts of memory crap as well as CPUs so just piling it on with more power might not merit the extra performance like a simple sha die.

In other news I am noticing that the FPGAs do get kind of warm when the board is sitting on a carpet. I could see those caps failing if they were not vented properly.

Hm. Are there any pinouts or protocol specs on either of their chips? It would be a hoot to put them on old jalapeno boards....

Fast update: Might have a Titan controller and dead units coming in, we'll see. If so I'll post pictures and updates on those, will be interesting to see what they look like.

Also put on the FPGA on the first board and now I have two running Neptune controllers. The secret to reflowing 100% pin BGA's is heat under the board, zephlux, and a really good watch loupe to verify the balls are on the pads.

Next up in the meantime is a weird Neptune board: This one has a short on one of the low voltage sides, but does not respond at all to a controller. It looks like it's hanging up the FPGA side of the SPI interface, might be shorted. So I'm going to pull the two bad power supplies and see if that clears it.

You should post the transaction ID and follow that happy little bitcoin through the economy. See how many other people it brings joy to....

Ok, back to topic: If you really want to do "free" power, figure out how you can make a miner help things. For example power factor correction. Could there be a way to use the inductors on the 12v to .6v switching circuits in every damn miner to smooth out the lead/lag loads on say HVAC equipment? How about non-linear loads? Utilities charge a dump truck worth of money to consumers that have crappy power factor, either put that extra energy into the inductors for the miner, or let the miner's fields dump into the lead side of the sine wave.

Then you mine bitcoins, make utilities happy, AND clients happy. They need smaller wires, they have less overheats, utility runs more efficiently, you get hailed as a fucking hero and get a statue.

Side note: I re-flowed a new FPGA on this other Neptune board... Test results below:

root@Neptune:/etc/init.d# io-pwr init
TPS65217 OK. Modification A, revision 1.1
root@Neptune:/etc/init.d# program-fpga spimux.rbf
dd: can't open 'spimux.rbf': No such file or directory
ERROR: CONFIG_DONE != 1 when nCONFIG is HIGH
root@Neptune:/etc/init.d# program-fpga /usr/bin/spimux.rbf
82+1 records in
82+1 records out
root@Neptune:/etc/init.d#

I think I can officially fix a rainy day. That's 2 for 2 on the controller boards.

And a few pictures. First, one with the Neptune power supplies removed.



And the water cooled Neptune, actually works great and allows the power supplies to breathe better...


Working on some other things for a bit, but still a lot of fun to be poking around in all of this.

Note: Board 1 fully operational. Power supply #2 runs a bit hotter than the others, that's probably why it blew the caps apart. New caps on, speed is 620gh at 270 watts with three dies running at full power (450) and the slower die at 350. Peak dc temp 78c, chip temp 45c.

So on to the supplies. Pictures in a little bit.

Ok, pulling the supplies sucks. I might get some cheat sauce if I do more of them. That said:

It's best to heat the board at full temps. Need to get the board over 150c to have a shot.

Flux the tops of the power supply pillars

Apply heat from the *top* of the power board.

Go 400c with the heat, moderate flow

Lift the supply lines first, then the back

Be prepared to lift off straight. Otherwise things on the back of the board will go flying.

That said, with both supplies off I still see a short. Which means either it's a trimmer cap on the board or the CPU is shorted. But there are 30 caps, you can't check them all...

However you *can* use an old trick of powering the short and looking for things that are warm. Warm things are probably the source of the short. It's usually only a few degrees C, but you can pick it up with an IR temp tool. Which I have.

However you can't apply a voltage significantly about the max voltage the component can take, otherwise if the part blows open everything else will fail big time. So I need a 1 volt supply at a thousand or so amps peak.

Fortunately I know exactly where to find such a voltage. Next week will be using explosive power to troubleshoot technology.

Ok, boards. They need more heat, need to be careful I don't accidentally re-flow the hashing chips, so I will put tinfoil over the underside of the chip. Kind of like how you keep your turkey from melting.

So anyway, board #1 in the screwed up world. Plugging it in with a Corsair heat sink on top (water cooling is so cool!) gave me a unit that came up but would barely hash. Checking the voltages showed a few things:

1) The layout of the power supplies is not easily apparent. It's actually:

2          0
3          1
4  5  6   7  
(Maybe, first shot at mapping them)
According to the code, power supply 3 was reading no current, no temp and power supply 2 was reading almost no voltage, 1 amp, and high temps (70c). Unusual. Sure enough the 1v rail normally will have a 30 ohm resistance cold, these two had a zero. Since one was trying to come up, I think the failure is in the other one shorting to ground and locking out the first one.

Great.

Now to figure out where it's shorting. These little power supplies are kind of cute: They are self-contained, can do all sorts of cool stuff, and have a pair of high side FETs on top and three low side FETs on the bottom. Low side carries a lot of current but has very low R(ds). High side is the opposite and they usually get hot as hell. So design makes sense.

On the low side rail short there are two places where it can happen:
1) Caps short
2) FETs on bottom side short.

Now to do some melting and testing...

On to the next problem, dead boards.

Taking a look at a Neptune hashing board I can see they have a big chip in the center which appears to be 4 separate dies in one package. Makes sense, as the design of this board is an implosion type, with 8 DC-DC power supplies around the chip and every two supplies power one die/side of the board. That way you don't have to schlep all the power from one point across the board. Wish everyone did it this way, oh well.

Anyway, board #1 has a nice brown discoloration under 1/4 of the die and sure enough that's where the problem is: The 1 volt line to the chips is shorted there, 20 or so ohms on the other three. My guess is a blown something, now to find what....

C

Ok, so let's see. First, here's a map of the KNC board with the locations of the main caps.



Next a picture of the running board with a reflowed TPS chip (check it out) and some of the caps removed.



Next we have what powers most of my work around here.



And finally the equal to the above pic in the re-work world. Seriously, a good preheater is the difference between using a blowtorch or a crem brule torch to warm your coffee. By bringing the board up to 200c or so you can quickly remove and reflow components with just a touch of hot air heat.

Thanks! Go for it, this is how we all learn and get better and stuff. I should do another talk at Defcon or something about this, need a good talk title (how to figure shit out when it's on fire? Hm...)

Well, sort of. On the hashing boards I will replace the filtering caps because they serve the purpose of both stabilizing the power input which is being whacked around by the DC-DC's, and because they can help in making the supply more efficient (power factor stuff, really interesting reads out there on that).

For the controller board, the caps are important, but a bit less so. You put them on the inputs for a similar reason but since the FPGA is only pulling .5a at 3.2000 volts on the input that's only 1.5 watts and only .001a on the 1.2 volt lines). So the exact values are a bit less critical and if you leave them off for testing purposes the world will not come to an end. So you can play fast and loose on these caps in the short term without destroying too much stuff. I left a few on (the most important one is the one next to the TPS65217 chip because that's where the DC-DC conversion and the chokes are) and for the rest I'll put them back on "later".

Finding out the values when the manufacturer doesn't give out schematics (boo!) is a bit complicated, but a $49 or so good Radio Schlock meter with the capacitance testing function is pretty good for getting close.

Now, if we were talking about caps in a RC circuit (for timing, checking waveform ripple across an inductor, or as part of a current sensing detector for an overloaded FET) then the values are more critical. But on the low power stuff on a Neptune board (aside from the fact that there is probably something like that around the TPS chip's regulator points) this is once again not too much of a problem.

Sure, I'll post pics of this reflow repair, and the times it took with heat to flow the components. Will grab the cam....