Sure, I can fix Titans, they're still worth fixing based on the prices on Ebay. Feel free to PM me about what you have.
|
|
|
Yeah, the little FET's on there are also very hot. IIRC when I looked through the tech docs on those DCDC's it shows where each DCDC's thermistors are located. I believer there is 2 per DCDC but I could be wrong. The docs also specified 2 key points of measurement for cooling & airflow, one of them was the inductor. Grilling through the tech docs is how I ended up realizing why in most cases these things are being tortured and the way we run them the life expectency is drastically minimized. Which is why I recommend 85-90C webgui readout being the absolute max we push them to.
They are, but the RC network across the inductor (that's how they sense current and stay in phase) is well tuned and I'm guessing the switching frequency is both dynamic and able to sense an impending cut-through with a crowbar of the gates to prevent total failure. The trick is you're supposed to provide airflow around the units, not just on the inductor. That airflow would help keep the caps cool. my guess is they are 105c rated and when it hits that the fun begins. Oh well.
|
|
|
I swear to God, I'm beginning to think I can fix a bad day. 6 BGA pads lifted, placed back by hand, new FPGA placed, aligned, flowed and the damn thing works. Yep, I definitely worked for this one.
I'll deal with the other three when the parts come in next week. But yep, I'm fairly good....
|
|
|
In the meantime I had a couple of interesting breakthroughs:
I know what the pins for those titans are now: They are 3.3 volts and 1.8 volt sources from the TMS power supply. They're not for the main power supplies (they have their own power sources) they might be for the EEPROM. I'm going to modify (cut) a 10 pin cable to bypass them and see what happens when I plug it into a controller:
If the LM75's come up with no power and it recognizes Titan that's not good. If they don't I will haul both and see.
I'll then try powering up the 12 volt line.
I think also I can see where an exploding power supply could take out cubes on a controller (and the controller). Basically that is a shared line to all cubes, if 12v got on those lines it would short anything there.
I just hope it ain't the dies. If so we're fucked.
Also I took apart the roaster board and sure enough both the FPGA and the TMS chip are shot. Replaced TMS, voltages are up, will think about putting my last FPGA chip on there. I have a several hundred dollar order on with Digi-Key, that will come in next week.
Next up: Titan work.
|
|
|
I don't think it's utter shit, technically there is always worse. Using off the shelf power supplies was actually a masterful idea since it took the risk of building a mongo power supply out of the equation. And I have no clue how they managed to get a high power scrypt chip together....
Then again the cases are hand cutting.... ok they suck.
(there's also the fact that building miners is a queen's race. I wouldn't do it)
|
|
|
Back to working on boards: Got three controllers in today, all dead. Bit odd in that they don't flash the green light which is what normally happens when the TMS chip or FPGA fails.
First board had literally blown two of the pads where the TMS chip is. Great. Put that aside.
Second chip fired up and lit the bright light, followed by green with no miners. Nice. Figured it might be ok, then it shut down the power supply. Not nice. Power cycled and felt temps. When the bright light lit, the FPGA started getting *exceptionally* hot. Hot enough to burn my thumb. Ow. Powered down, went to pull the FPGA chip (bad).
FPGA came off, but some of the traces underneath it literally had delaminated. Wow. Now it's true that the controller board is not designed for heat, but this is a first. So I cleaned it up, took off the solder, and straightened the 6 pads that had come loose. Tomorrow I'll sit down and see if I can put a new chip on it; the balls should pick up the pads and provide stability.
This however points to a failure mode: FPGA degrades and pulls more current. Gets hotter, then either the FPGA shorts or it blows up the TPS chip. Or in this case screws everything up.
Never dull.
|
|
|
That is a really, really smart idea. Covering those inductors was one of the more brainless things I have seen.
What do you mean by "covering"? ... the stock plate? Cuz that stock plate actually functions as a single large heatsink ... maybe a poor one but it still does its job. Mrph. Most of the heat can actually be felt *under* the board which is the wrong place for it to be. I haven't actually gotten a temp probe to play, but in almost all cases the real source of heat is not the inductor it's the high side FETs. And those get no air on the 1/2,3/4 supplies. And those are what is right next to the caps that usually blow sky high....
|
|
|
That is a really, really smart idea. Covering those inductors was one of the more brainless things I have seen.
|
|
|
Sounds good. Need to place an order with Digi-Key for some more FPGAs and power chips. And just sent out a repaired Titan, we're hitting a stride here.
|
|
|
lightfoot,
Need you address to send in my sick controller. Changing bridge and RPi didn't work. Controller still restarts every 2-3 mins, doesn't recognize any cube, LCD screen stays dimly lit w/ no bright LED or any LED.
Yep, short, sure, send it in.
|
|
|
ASIC 31 appears to be shot. Interesting.
|
|
|
I received 2 replacements for dead boards that I sent to Bitmain. Within a couple hours one of the boards is dead again... What a waste of $ and time. I know their response is going to be for me to ship it to them again for repair, because they did so well at gaining my confidence the first time...
Oh man, that sucks Are you using their power supply? No. I use EVGA PSUs. I've got several S7s running and the others have had no issues over a several month period. Man that sucks, they should have someone fixing these things stateside. What seems to be wrong with it?
|
|
|
Ok, so on both the Neptune and Titan pins 2 and 8 go to pin 1 on U14 and U16. Which are an LM75 (temp sensor) and a 24C32WP (Serial Eeprom).
Well, it's not pins 4 and 6 but it's a start. Hm. What were you doing KNC?
|
|
|
Ok, that is *very* different, and it might be good to modify your previous post. My belief is that information should be free, but ability should be paid for properly. In my case, I post the things I find here as a service to the community, but I charge for the work I perform. If someone else can also do the work then all to the better as we get more home miners.
In Tarkin's case, he took the time to write the code and you can either pay $50 for it (cheap as hell) or pay 1% mining fee (also cheap as hell). When I get the Titans running on Beaglebones I could probably charge as well, but to be honest I'll just post the basic steps. If someone wants me to reprogram their BB then I'll charge some bitcoin amount to do it.
Either way far cheaper than either resetting your miner all the time (Tarkin) or spending $1k for a Titan controller (me)
|
|
|
I do know about Tarkin but that software takes over the miner once every day and hashes for tarken for about 11 minutes. I am not comfortable with that. I don't mind paying the 50 bucks but I think its a bit greedy to keep extracting payments everyday indefinitely thereafter.
WHAT? Really? First I have heard of that.
|
|
|
Don't know. Tarkin has this all figured out, $50 is cheap for the work he did.
On a related note I know why my last two Neptunes are not working: The boards are literally curved like a taco chip. Apparently someone in the past tightened the screws down so hard the heat sink warped the board. Gets hot, connections lift, bye bye board.....
|
|
|
The controller doesn't share a common ground with any of the cubes correct? So you are allowed to plug them into different breakers?
Well, hm. The controller has a common ground line to every port, the supplies to the FPGA are not transformer isloated so each cube does share a common. One could have used optoisolators on the SPI busses, but naah. That said, I don't think different breakers would be a problem as long as all breaker and power supplies shared a common ground.
|
|
|
Interesting Neptune here: This one seems to have two dead supplies. https://www.flickr.com/photos/137949766@N06/shares/6Hs3ESDC/DC Voltage (V) Current (A) Power (W) Temperature (°C) 0 0.7839 37.0625 29.053 80.375 1 0.7826 37.6875 29.494 91.125 2 0.7809 39.0000 30.455 64.625 3 0 0 0.000 0.000 4 0.7819 37.3750 29.224 56.938 5 0.7826 38.0625 29.788 65.000 6 0 0 0.000 0.000 7 0.8092 0.0000 0.000 42.438 And hot on the bottom. Fair enough, standard bad connection/etc. Opened it up, and *nope* It's was better. Oh yeah. Someone had torqued the screws down so hard they literally *broke* the aluminum base. Yes, another one was broken too. Yes, nothing was holding down the heat sink really. But was it KNC? Nope. That's not their heat sink sauce. So this unit was basically running without a heat sink and probably imploded two supplies. I'm going to declare this box a wreck, and see if I can get the board up on another box. Oh well, at least I know that you can't tighten the screws enough to break the chip. Moral: Don't tighten the screws so much.....
|
|
|
Yup. Those caps short, and sometimes incinerate the whole power supply. I just fixed a Titan where two supplies blew up, one so badly the traces under the board were wrecked. Fortunately it can run that core at 150mhz on one supply now that everything is cleaned.
Just checked it, at 500gh/350mhz temps are 45c chip and 69c power supplies. Fine.
|
|
|
Ok, so an update on that super-hot Neptune (Titans will do this too). Thing was running with PS temps at >100c, chip at 60c+ at a mere 350mhz, bottom was extremely hot to the touch (these things must be laughs on carpet). Took unit apart, found this: Yep. Filth and crummy hardened heat sink compound. Cleaned up the chip, the box, everything, and now: 44C 0 0.7838 34.5625 27.090 75.500 1 0.7834 35.0000 27.419 79.625 2 0.7859 35.0000 27.507 67.625 3 0.7823 34.3750 26.892 67.750 4 0.7838 35.3125 27.678 60.000 5 0.7870 35.7500 28.135 65.625 6 0.7848 34.2500 26.879 63.688 7 0.7828 35.0000 27.398 65.125 500gh, 218 watts, bottom normal temp to the touch. Moral: If your unit is running hot, try opening it up and cleaning the thing. Replace the heat sink compound, clean the old crap off the chip (if there is hardened crud on there you will break the chip when you torque down the sink) and torque the screws down "finger snug". Likewise snug them in sequence, to even the pressure. Never dull, on to the next unit.
|
|
|
|