Hacking KNC Titan / Jupiter / Neptune miners back to life. Why not?

[Fixx]

(ie moral support)

Thank you so much! )

On the subject, it seems flash memory on the motherboard miners are problem

It looks like I now have a very expensive brick

[lightfoot]

(ie moral support)

Thank you so much! )

On the subject, it seems flash memory on the motherboard miners are problem

It looks like I now have a very expensive brick

On the Raspberry pi? If it's crap, just buy a new one on Amazon. Make SURE you get the 1.2 version Revision B and not the newer Raspbery Pi II or III.

This one should be it:

https://www.raspberrypi.org/products/model-b-plus/

[lightfoot]

First day in the mid-80's (F) here, and wanted to let you know my Titan control cube here is running at 56c on the die, 90c+ on the power supplies. 300/325/325/325. I was running it full blast with two 750w Corsair M power supplies feeding the double plug to see if it would overheat the Molex. (oddly enough it did not).

NOTE: When it gets warmer the air cooled units will run a lot hotter. This is 20c hotter than it was a week ago with temps in the 30's. Be warned, I am clocking down to 250mhz.

[lightfoot]

Hm. Clocking down to 250 did not work, two sets of power supplies dropped off. Switching voltages didn't help either, below a specific clock speed (275) dies will fall off line.

Which makes *no* sense, but seems to fit a pattern: Titans have a sweet spot where they are stable in a voltage/frequency level. Run them there, things are good. Run them other places and the power supplies destabilize.

[joef]

Hm. Clocking down to 250 did not work, two sets of power supplies dropped off. Switching voltages didn't help either, below a specific clock speed (275) dies will fall off line.

Which makes *no* sense, but seems to fit a pattern: Titans have a sweet spot where they are stable in a voltage/frequency level. Run them there, things are good. Run them other places and the power supplies destabilize.

They start to behave like that, when they get older. Maybe it helps to replace the 5 ceramic capacitors at the input of the power supply. These capacitors are dedicated to reduce interferences triggered by the ASIC power consumption changes. With a 300 MHz oscilloscope that should be visible.
Did you found out the capacity of those 5 capacitors? In one of your previous posts, you wrote something about 25 uF. Are all 5 identical with 25 uF ?

[qberty]

hey lightfoot, sentcha a PM.

[mc162BCT]

Okay but how do I cut the voltages?

Try -.06 volts instead of -.03 volts. Then try 400mhz speeds. See what you get.

C

An update...https://i.imgur.com/aayIBQA.jpg


I allowed my 3 cubes to go grab some extra gigahashes and by sunday evening I began to smell something a little "off" however everything continued to mining so I ignored it like a dude who had cubes to burn so to speak. 

In fact I actually did have one that was free or late shipping (ebay) so I figured letting just that ONE run at or approaching "StoCK" settings like 450mhz-.366 pulling 270 watts was kind of a gamble that I consciously decided to take and sure enough that smell was a harbinger of the inevitable probably was too late by the time they can produce that smell anyway but so yeah

Here's the autopsy photo's.

https://i.imgur.com/nvP6nSx.jpg

https://i.imgur.com/MkhXWvs.jpg





PS..Pardon my noob-itude but how does one post photo's/jpgs etc directly into threads/replys?

I know there's a way but I don't know it obviously...anyway if somebody might have a suggestion on that I'd appreciate it.
Apologies in advance but 2yrs lurking on and off with this place still haven't a clue how a vast majority of any of this stuff works...

[lightfoot]

Ug. Yep, how many watts was the cube pulling?

[NotFuzzyWarm]

Instead of pasting in the link as a hyperlink, select Insert Image which is the icon on the left of Insert Hyperlink. pasting the link info into that results in


A tip - if possible please crop the image to just the area of interest to speed up loading time on slow connections...

[irritant]

no he cant because he is a noob
(not enough posts)

[Searing]

I've been using Corsair 1200i's on my Titan setup. My Original Corsair on Original Titan lasted 13 months or so. It just kinda would fade off 1x or 2x a week.

So Corsair let me send it in as a return. The unit for my 2nd Titan March 2015 is still working fine.

So I get the replacement Corsair 1200i ...looks new...box new...unsure if it was really refurbished or just replaced...different serial numbers etc

And it has been around 4 months and now it is fading also with only 2 cubes (i've added) since ...so its not load nor heat.

Is this just the common way PSU's kinda die.....? Using Gen Tarkin software so there are occasional soft reboots etc .....is the only real change.


Anyway just wondering if Corsair 1200i's are known for this kinda thing and I should just take the next replacement (likely they will do so) and put it on

a shelf and use a different branded platinum PSU etc etc


Not a good sign...1st one 14 months ...acceptable.....2nd one I got 1 year 1 month still going strong...the replacement I got on RMA 4 months ago

now fading on a different controller and cube setup in the same manner as the first unit I sent back and this is the replacement for that


Anyway ...maybe it is  a 'true' refurb with a cursory check at Corsair's Service Dept end and this is common....


anyway just tossing it out there for those with BIG farms that may run into this kinda thing more due to more units or something

see link below for my setup ..the miner pic shows the 1st titan with 6 cubes (stable as a rock go figure) the 2nd Titan shown setup is same but only 2 cubes now on this

fading 1200i corsair....ie heat is NOT and issue at all (yet)

www.lostgonzo.imgur.com


anyway again wondering ..the dance is getting kinda old ......maybe on fixed/returned/rma units this is just the way stuff shakes out?

[hinvestmentgroup]

I've been running my two Corsair RM850 since October 2014 without any issues yet.  Of course now that I've said that I'm sure they'll start to go.

[Searing]

I've been running my two Corsair RM850 since October 2014 without any issues yet.  Of course now that I've said that I'm sure they'll start to go.

my setup

lostgonzo.imgur.com

I am running 3 cubes to a corsair thou FYI as you can see on the above...perhaps even for a 1200 that is pushing it maybe..thou the one that seems to have failed

2x in last week..now has one cube on it lol .....only ever had 2 cubes on it.....hell its bad no doubt i guess


Yeah 1st one went after 14 months or so

Have 2nd one still going after 13 months

this 3rd replaced by corsair one ...4 months sheesh...and unhappy trend

the one that is fading now is the RMA new (or refurbished) can't tell replacement for the 14 month one that went

on a side note my black/yellow made molex psu 3 lead y adatper plugs all started to burn out after 14 months ..replaces almost all of them..the KNC ones had 1 burn out

the other seems fine.....knc that is ...so far..

the other couple non knc  seem  unaffected so not  replaced yet...but that is also a consideration.....not a slam on the made plugs i had made...but something to consider

stuff like that may also start to go at the 15 month or so mark.....also would depend on how much 'pitting' you'd get from moving cubes around and such

I think I will have no problem sending the corsair 1200i back its only been 4 months ...but hell the stuff is in the basement and not even warm yet

i dread summer


have a buddy with extra platinum server psu's that can be mod'd to work 95% eff they are 1250's etc 100 bucks each I'll try instead ...maybe if i get another corsair 1200i

back new(ish) as a replacement yet again I should just keep it in the box for a swap out replacement and hopefully the above server psu's will do the trick at less cost

(and hopefully this yo/yo will stop)

[lightfoot]

I run my sample cube with a pair of Corsair 750M's, one per side to a single Titan. With that the connector does not overheat at 300/325mhz per die. A single supply with one cable will overheat the Corsair plug. With two cables it will overheat the connector on the Titan due to voltage drop. With two supplies, two cables the voltage is high enough that the amps on the cube's 12v side don't burn pins.

Really, be careful. I have another board here with a burned plug, and a board came in that had the sink torqued down so hard it CRUSHED THE BOARD. I literally am going to have to take some pictures of that. As in ripped open on the bottom :-)

Back to work.

[lightfoot]

Well, this is a step forward: I managed to get the top off a die without destroying the carrier.

The dies are larger on this board, I think KNC did two revs:



More interesting, someone tried to fix this board. Bad job!



The secret is to use full board heat, then heat around the side of the chip, then very very gentle pressure to lever up the lid. Bitch on wheels to do, but in this case the top is off.

Hm.

[BitcoinPappi]

Well, this is a step forward: I managed to get the top off a die without destroying the carrier.

Is this good news for -> ?

Titans, but some of the issues, especially the ones where the Titan shorts the controller are out of reach. Vegas, do you have a stencil to reball a Titan die? The big problem with them is related to the 1.2 volt lines to the die controller shorting out, if you can remove those pads or cut the traces under the board you would win big time.

[lightfoot]

Is this good news for -> ?

It's interesting, and a solid day's worth of work. However thanks to someone sending over a literal taco board, I was able to work the lid off with the right amount of heat without breaking the chip free.

I'll work on it more next weekend, need to see if I can do this on a Neptune, then run the unit without the lid (sinking directly to the dies). Ifso I can try lobotomizing a die and see if I can clear the solder and get a nep to work. If so then I try someone's volunteer.

[lightfoot]

On a related note, I *HATE* replacing TPS chips. And note that a Rpi can short out the CPU chip which puts 5 and 12 volts on the lines to the TPS chip and FPGA. The resulting damage sucks to repair.

But sank the day today fixing this board. It's up and running, now the owner just needs to send over money for a Rpi, I'll order a new one, and send it on back....

C

[GenTarkin]

Well, this is a step forward: I managed to get the top off a die without destroying the carrier.

The dies are larger on this board, I think KNC did two revs:



More interesting, someone tried to fix this board. Bad job!



The secret is to use full board heat, then heat around the side of the chip, then very very gentle pressure to lever up the lid. Bitch on wheels to do, but in this case the top is off.

Hm.

Wait, WHAT?!?!? the dies shown here and the dies shown a couple pages back on forum ... are really really really freaking different in size! Are these both Titans?
If so, it looks almost like they could have used 2 different node sizes. Its such a substantial difference in size the revision would have been an insane undertaking. This is weird.

If these 2 pics are different products tho (IE: neptune vs titan) then such differences in size are normal.

[Searing]

Well on another note.. the Gen Tarkin Software went ACK! and shut another die to OFF..on my orig dead die I got from knc so now down 2 dies..

the Gen Tarken Software .goes all SKYNET on me when I try to tweak it ...heat issue ..do not pass go...do not muck with the die....heh not a good sign

when you try to turn a die on and reboot and during reboot slap'd down and put back to OFF after like 15 seconds or so lol Tarky Titan Software NOT amused



So......I guess I should re-paste stuff and add the heatsinks better fan etc (stuff from the Swedish guy's video and his parts)

So I guess it has been for 4 of the cubes on my 6 cube unit been 1 week to go be 18 months..the consensus of Maxumark is my paste is turning to ASIC

faerie dust as we speak....so summer heat is coming....break it now trying to fix or burn it out later (its like Sophie's Choice for ASIC dies)



Ack! Guess I'll have to "MAN UP" and just try to start the 1 die only working cube.....its one step away from paperweight anyway...(this could end badly)

I would imagine there are a bunch of folk with NOV 2014 or so Titans that with Summer coming are also going to learn the joys of 're-paste and heatsinks' on Titan cubes.

Good to have goals