Agreed. In a house (which doesn't move around much) there is no better bang for the buck than either a T105 or L16 based bank.

That said I run 12-370's and 12-400's in my shed because I was too lazy to build a battery box to vent them properly. Don't be like me.

Tollip: Nice job! What I did was land the +12 from the caps on Supply 1, through the +12 adapter's 3 pins, then on to the other supplies like so.



Works well enough, especially when you consider the damage on the pads from burning out the plug.

As for pushing and pulling air, when you push a fluid or such into a load (such as a heat sink) what you get is back-pressure which can cause the fan blades to stall. Once they stall, the fan will spin faster because it's not doing anything anymore. By pulling the air, you increase the velocity around blockages (like fins, radiators, parts, stuff like that) resulting in more airflow, less stalling, and lower temps. I can publish some numbers (since I'm a numbers nut and have a lot of junk here) if people are interested.

Had to rebuild another board with a burned +12 supply line. One pin was gone, one pin had blown the inside open, the third pin had 20 ohms of resistance. Not too great.

Up and running right now at 200mhz, purrs along. Tomorrow the new Rpi comes in, so we can test if the transfer boards can be fixed. Should be do-able, just need to find my wire-wrap wire.....

Yes.

I suppose. One of the things to watch out for is that you would measure incoming air temp as the KNC titan pushes air through the enclosure. Technically it's always better to PULL a fluid (ie: air) not push it, but once again running full steam is best on a Titan that is still quite worth something.

Meantime I'm wiring another bypass circuit up for someone who torched a cube's power bus.

Pretty much. You need a resistor bank or a dc-dc controller and to be honest it's really not worth it. Besides, the fan is pretty quiet as is, and the board has very limited temperature sensing so run at full speed.

A what?

Not a problem, pleasure to fix and fun to do this for bitcoins!

Not sure. Why would you want to go to 1.00? 1.06 runs fine.

Eh, not that bad. To do it right you need to remove one of the sockets with air heat, then clear the fault and run a nice 24 gauge wire-wrap wire between that post and pin 3 on the bottom pin set. But you could just glue down the bad bits and jumper directly with said wire if you're in a hurry. And who isn't in a hurry.

But you need to keep that burned line secured with glue or something in the center of the connector, otherwise it is going to short to a gpio line and all hell *will* break loose.

Well, here is how the Titan bridge boards blow up:



I can fix that More to the point the failure chain is this:

Raspberry Pi shorts out it's internal DC-DC converter circuits.

Results in high resistance on +5/gnd on pins 1/2, but as soon as power is applied the board shorts as it tries to bring up the low voltage supplies.

Big +5 supply from the controller board shorts, blows up via line.

Fixable.

And in the "Take some pictures of what's up" mode, we have the following:


This is the connector on a display module that had some of the pins not soldered properly so it worked intermittently. Notice the tilt. Bit of love with a soldering iron and some flux and all is well again.


A bad repair job that was sent to me. Note the bottom right pin on the molex and how it is distended. That via is blown clear of the board, so this board would only use 2 +12v pins. Note the black mess. That is due to using a big soldering iron without pre-heat. Sad. However this is the one where the chip core is shorted to +hashing power instead of fucking ground, so there is a chance of a snowball in hell that I can figure this out.


A nice old Jupiter controller board retrofitted with two new 10 pin plugs to even it out. Take your time, desolder all the holes, then use hot air to allow the plugs to come through without pressure.


Yet another reason not to push cubes to the limit. That is a stock KNC splitter, the problem is the damn plug isn't rated for that kind of power draw....

Now I have a Titan controller on the bench that blew up. Oddly enough looks to be the Rpi went bad as the controller baseboard is happily hashing with a bbb board on it.

Try running all dies at 50mhz, then 100mhz. See what happens then.

I suppose. However when someone puts a cube on your doorstep that did this:




You wonder if that cube plug is just a limiting factor.

Ultimately it's up to everyone to do what is best in their circumstances. :-)

Well, these two units are currently running approximately one (1) inch away from each other, so the environment is pretty constant. The difference between the two is the difference between an average DC-DC temp of

ASIC slot   Temperature   DC/DC avg temp   Clock   Type
1   ---   ---   ---   OFF
2   32.5 ℃   47.4 ℃   225 MHz   TI
3   ---   ---   ---   OFF
4   ---   ---   ---   OFF
5   ---   ---   ---   OFF
6   41 ℃   62.0 ℃   275 MHz   TI

Which is pretty big for that extra 10mh.

You mentioned that your power supply cables melted at least once, and I just saw a picture of a guy's power supply that basically started to set fire to his house. Glad his house did not burn down, but the problem was the same, power supply cables melted. Fixing melted units that blow the PCB connectors is a real bear and I just got a Neptune in that had a blown power cable that led to a short on pin 6 (this time to the .7 line which allows it to work, but very weird) which proves board burns can damage other Neptunes and Titans due to the lack of signal isolation on the controllers and/or the units.

Perhaps a good code release would be a "do not pull more than X watts from the supply lines" so people could put in 180 to be nice and comfortable up to 260+ if they really want to see things go foom or something. As always I stand by to fix the blown cubes, but it would be better if they didn't blow in the first place.

Some hard data for you Titan users: Moral: These things were built to hash at 60mh, going faster makes miner sad.

I'm running two miners here, one is a stock reference one running at 275mhz, the other is a client's repaired one (with bypass wires) running at 225mhz. Note the power, DC-DC temp difference, and other differences. It's a lot of extra oomph to get that magical extra 10mh....

ASIC 2 (Speed 225mhz)
Temp : 31.0 °C
Power : 187.445 W

DC/DC   Voltage (V)   Current (A)   Power (W)   Temperature (°C)
0   0.7878   28.781222.674   47.500
1   0.7874   29.1875   22.982   46.900
2   0.7886   29.4062   23.190   45.900
3   0.7883   29.9062   23.575   45.300
4   0.7920   30.0938   23.834   44.100
5   0.7922   29.3438   23.246   43.900
6   0.7892   30.3750   23.972   43.900
7   0.7892   30.3750   23.972   45.100

ASIC 6 (275mhz)
Temp : 38.5 °C
Power : 232.520 W

DC/DC   Voltage (V)   Current (A)   Power (W)   Temperature (°C)
0   0.7858   36.5000   28.682   61.600
1   0.7863   36.5625   28.749   63.400
2   0.7819   36.8750   28.833   61.400
3   0.7841   36.9375   28.963   60.400
4   0.7836   37.3125   29.238   55.900
5   0.7847   37.9375   29.770   55.700
6   0.7826   37.3125   29.201   57.700
7   0.7861   37.0000   29.086   57.100

 KNC 0:       | 59.00/61.52/61.57Mh/s | A:315 R:1+0(.32%) HW: 64/.59%
 KNC 1:       | 71.02/72.75/72.66Mh/s | A:385 R:2+0(.52%) HW: 73/.57%

I would recommend slowing down to 60mh total. That is what these were designed for.

And while 571 is prime, 63*9 is 567 which leaves four dies left over. So they could have 9 strings of 63 per die. Hm And once again it would REALLY be helpful to have ANY specs on these things....

That's interesting. 571 (2284/4) is a prime number.

Why would they ever have a prime number of cores per die?

Hm.

Ok, here's the thought: Pin 6 is power to the chip's housekeeping circuits. Each die has capacitors on the bottom that go to ground and to this common line 6. The line comes out to a pad on the bottom, then goes via two small traces to two pins on the chip. There are 5 pads per die, the outside two power two pins each, and the inside three power 5 pins together making 9 connections to the chip.

This is different from the actual engine power, which is supplied by the big DC-DC's. Think of this power as what runs the signal thing that polls the engines, puts work in the engine memory, that sort of thing. Each housekeeper is responsible for some number of engines, and they all then communicate on the low voltage spi bus which goes back to the fpga for figuring out.

What I think happens is that when the chip shorts due to a bad surge everything welds to ground on that die. die power, as well as the engine power. This is what we see when a board burns. Worse is if one board loses its' ground, then the ground gets pulled over the lines from the other chips which makes a real mess of either burned ribbon cables or blown dies.

So what to do?

Well, isolating the dead die is the right thing. Since there exists no stencil on earth for us to re-ball the die and remove the balls for the blown connections (the simple solution) we have to cut the lines.

The common line for those dies is very thick and runs all along the back of the board. However with a very sharp knife we should be able to cut the 9 traces to the pin vias. I'm going to need to order some very sharp surgical knives with fine points; copper dulls steel quickly and this will require very precise cuts so as not to hit the layer underneath.

Hm.

What is this "block" you speak of?

Question: Does anyone know how many engines are in each core of a Titan? Got a theory here....