Maybe. I have one incinerated Y adapter, showing that even that can't make up for a bad power supply or overloading or a connector that has been plugged and unplugged with the power on (thus causing sparks which cause little bits of resistance which warm up which.....)

Still, it is also true that there seem to be these islands of stability on Titans, and if that island is at 300-325mhz then you're kind of stuck with it.

Back to work here. Have two more boards to keep me busy over the next few days.

.0336v, 325mhz is a tad bit... much.

Temps that cause the power supplies to go over 85c

Stuff like that.

There's a pretty big power difference between 250-275 and 325. Bit less power seems to result in less problems.

C

Meantime I got these pictures in. This is a Titan with both jumpers in place:



I know this had all cores running at full blast because you can't set a power plug on fire like this...



With anything less than a full component of hashing power. :-)

Side note: Do you really need to run at full power? 275mhz isn't that bad....

It's a 0 ohm jumper. Usually you put those in if you want to isolate something during testing or production. You can also see some around the Titan chip itself, look for a pair of traces that have a jumper, then a jumper on the trace next to them. That I think is the signal lines to the chip, those other jumpers we're talking about go to a 3.3v vcc2 supply from the controller board, common bussed.

It's never as easy as we wish :-)

Well this is a bit of a bummer. Basically what I think I'm looking at here is binned chips. More specifically, those jumpers supply power to the control circuits on the hashing chip on either die 0/1 or 2/3. Along with those four sets of jumper/resistors around the board that's how the Titan talks to the control aspects of dies 0,1,2,3.

With the resistor out those dies are off. Put them in and the die controller is powered. At that point those dies now talk on the chip bus, and can chatter/jam the bus if bad. So KNC never put the jumper in if the chip had bad dies on the top or bottom, they probably sold these as the "filler" to make out a set (so 3 90mh Titans plus one 40mh would still make a fair 300mh or whatever it was they sold this as).

Put the jumpers in, turn on the supplies, and the chip can run with all dies. Or it can go insane. Or it can run one die and not the other. Or it can happily run at 100mhz but jam the other die at 150. Strange things happen, and what I am *really* wondering is if this means that General Tarkin's code can fix this by doing his sweep the frequencies and reset a die if it goes bad (the "warm restart" thing). In which case he's going to be a God. We'll see.

Hey, we have to acknowledge failures as well as successes. It's the only way we learn. I don't charge for failures, so if things don't work there's either no charge or the bitcoin gets sent back.

Onward to the next thing. :-)

Bad news cube ran file all evening but after letting it sit for 4 hours dies 1 and 2 throw errors. This is why you have to test your work several times.

And why we acknowledge failures as well as successes.

Crap as they say back to drawing board. And it was hashi g so nicely.....

CRAP! Short summary, may be unstable after powering down and up. There may be a reason they disabled the dies. That's why testing includes powering everything down for 4 hours then a restart and retest before shipping :-)

Ug. Crap crap crap crap crap.

Welcome to Titan land!

It's probably a shorted FET on the high side of one of the power supplies or a blown apart capacitor shorting things. Figuring out which FET it is takes some time to do. Problem is a short through on these Titans usually takes out one of the dies. However clearing it gets the rest online.

C

Possibly. Send one first, let me see if this fix fixes that.

However if you have two dead dies and it's been that way for a long time let me know.

Or if your Titan blows power supplies when plugged in.

Never dull. I'm going to bed, will see if this thing is hashing in the morning. Keeping the speeds down, but even 250mhz is purring along at 65mh.

That would explain all the orphaned shares I have been pulling as of late. Off to fix bitcoin.conf again...

And on a non-related note, I just figured out how to fix a Titan cube with 2 dead dies to hash at full speed.

It's Miller time!



Lightfoot

Well fuck me......................................

DC/DC   Voltage (V)   Current (A)   Power (W)   Temperature (°C)
0   0.7863   34.6875   27.275   62.200
1   0.7810   35.3750   27.628   65.100
2   0.7867   35.5625   27.977   66.000
3   0.7897   35.5000   28.034   61.900
4   0.7866   32.6875   25.712   56.200
5   0.7850   33.3750   26.199   57.900
6   0.7885   31.8438   25.109   55.900
7   0.7886   31.9688   25.211   56.900

KNC 0:       | 76.06/72.23/72.67Mh/s | A:12 R:0+0(none) HW:3/.64%
KNC 1:       | 84.17/80.06/80.37Mh/s | A:13 R:0+0(none) HW:3/.55%

Ladies and gentlemen,
Boys and girls...
Party time is here......

https://youtu.be/s8MDNFaGfT4

Well those are worth something :-) If you dare take them apart and post a picture of each board at high resolution so we can try to compare what is what and where is where.

And never plug a suspect Titan into a working string of them without testing first :-)

I'm wondering if these were binned from the factory. They promise 300mh per four cubes, 3 cubes with 4 dies, one with two. Or something like that. The missing jumpers is kind of odd, they go somewhere.

Meantime I'm back to square 1.5 with the blown titan board, I hooked up a second engine and the thing now shuts down the supply with pin 8 connected agian. Back to drawing board another day, I need to cook some food.

anyone else got jumper news?

I've seen it twice where one of the two supplies dies and the unit is not shorted. The issue is the remaining supply tries to power the die and if more than 50% total hash is requested supply 2 will crowbar and shut down. So I try those at 1/2 speed first (well, 100mhz) then step up.

Replacing the power supply is possible, but to be honest such a pain it's not worth it if you can get 50% hashing.

Sunday morning work: Pin 4 has been identified which fixes the issue on the +3.3v supply going to power the LM75, EEPROM, and one of the level converter buffers. Will clean up the board in a bit.

The other is an issue: Pin 6 is still shorted and on the first board was shorted hard. It powers the other 3 converter chips, and I just found out this morning (by looking at a scrapped Neptune board which I had lifted the chip) that it also goes to the hashing chip on eight separate pins. So that's the supply for the housekeeping circuits on the hashing engines. Mystery solved.

The question is what to do about it: I can try forcing up to 5 amps of current to see if I can get the line to 3.3v, letting the shorted die roll in which case I need to put everything back together here. I'll try that this morning now that I have pin 4 and 8 repaired. That might be enough to get at least one engine running, which would give me a foundation to test other things.

Holding off on other work on the Titan with the missing jumper for now, I want to see if I can trace out that specific line on a Neptune first.

For Asic 2 it looks like one of the two supplies is out. Try running it at 100-200mh and see if it adds to your hashing.

For Asic5, that's the question we're trying to figure out.

Close. I'm not in Hawaii (that would be cool), the Post Office is just stupid sometimes....

Full stop here, found something odd.

As we know, these blown out Titans have pins 4,6,8 shorted hard to ground. Pin 8 we know now goes to four points on the hashing chip, my guess is that it's some sort of signal or power.



Pin 4 is also pretty simple now, it goes to the temp sensor, USB, one of the buffer chips, and to this odd spot on the board.



Pin 6 is the one I am tracking down now, I know it goes to another 3 buffer chips for power but it also seems to get lost in the board....

HOWEVER this is one guy's Titan that only hashes on two dies out of four.



Note the difference? Note between U9 and U10 there are supposed to be two zero ohm resistors and a cap. However on this guy's board there is only one resistor, a cap, and an open bridge. With nothing soldered onto it.

Why? Question: Can people who have two downed dies check this area and photograph? Because on these blown out boards of mine, that same spot where the resistor is missing is the ZERO OHM SHORT on pin 4.....

Something is up here.

mm: That's a nice thing. Even if they said what the 10 pins did it would be a plus.