Topic: Hacking KNC Titan / Jupiter / Neptune miners back to life. Why not? - page 22. (Read 76793 times)

As stated in PM ....anyone know who made these chips for KNC in China? They likely would give you the stencil/grid thing you want to refloat the die (man above my head)
anyway you know what I mean.

Only other thing is if you have a local University with an Electrical Engineering Dept (and say EE club) and/or a Vo-Tech for Electronics Tech Training...and ask around they
may help you for the fun of it and beer and pizza....kinda learning real world hack they may get into even at the faculty level

added:

I have a univ EE dept here..if you are really really specfic I could ask around for such ....but probably better you ask in person someplace near you....but hey I'm game
if that EE thing is not an option in your location..worse they will do is say no.

One other thing ...Qberty has connections to make those Clone Titan Bridges...maybe the folk he used to make them for us would have an angle on this ..anyway all I got

Ok, back to the drawing board. I now have a dremel 4000, I can't use the jeweler's gravers because the board is way too big to fit in the staking tool. Oh well. So what's left?

We know the chip/die is shorted internally.
We know KNC bussed the connections internally to the board (fuck).
We know the vias are 1mm in size.

So I'll go in deeper and order two key items:

1) Drill press for dremel tool. Not as good as the staking tool but should get me there.
2) Some micro-bits for drilling pcb's.

If I can't use the dremel I'll use the big drill press. Goal will be to drill out the vias on the shorted chip side.

Unless I can get a massive bga stencil, this is the only idea I've got :-)

Update in the not-so-good news category:

I picked up a high power dremel tool and a milling bit in order to finally isolate the die hotel power line that gets shorted from the rest of the dies. This should be do-able by cutting the double and one single traces from the capacitors under the board to the vias for the pads into the chip. Very tough to do with a knife because of the rather thick copper foil used on the board.

However a mill bit will work. Cut the traces checked the resistance, still zero. Checked a blank board to see what the resistance is, it's zero as well, which means those lines are common on the board and can't be separated out.

Why oh why did they do this? It really makes no sense in the world, but it means in practice that there's no way to isolate that power supply line from the rest of the dies. Which means boards that die with shorted pin 4 or 6 are dead.

Drat. Ok, back to the drawing board. Anyone know if I can buy the schematics for these boards from the bankrupt entity that is KNC? I'll offer them a litecoin.... :-)

I would just make all the pins bare and do what the pic below shows...what I did recently...works slick nice tight fit ..no issues at all

Below is Maxumark hack that works just great for me

I just 2 days ago had a plug blow out for the first time on a cube....so I did the cut out all the plastic from the pcie plug and slid on a knc Y adapter as shown in pic below
(see link also)

https://bitcointalksearch.org/topic/m.13872248




I'm also repasting the cubes as I go along they are dirty dirty dirty....yech....oil from those crummy rubber pads on the heat sinks...yech....so motivated now to do this


anyway the above hack works just fine ..the cube is back at 325mh and 10c cooler to boot

FOLKS YOU HAVE TO GET those fans from the Swedish guy video

these (also got his heatsinks from his source) are beyond frigging great!

The difference between the crummy 'no name' fans that KNC labeled (orig batch) no way are those fans 2k in speed the below 3k fans blow easily 2x the air an hell

they are barely louder then the crummy knc fans..huge.huge diff imho

here is the link on newegg

http://www.newegg.com/Product/Product.aspx?Item=9SIA9PV3Y64442&cm_re=nfa-14_3000-_-35-608-048-_-Product


using Maxumark's X bracket mod and the tall heatsinks complete the fix...man does it work well ...I think they were like 10 bucks! again clean clean fast airflow with fan and this

https://bitcointalksearch.org/topic/m.14438879



but again the fix above uses the knc y adapter hack to the cube...and it is working well for me

Got to love how the community has come together and kept these knc '90 day wonders' up for all this time.......the X brackets the firmware mods by gen tarkin the hack above

whatever floats the boat...and 5.51 LTC sure don't hurt either (at least we 'outlasted' knc with their bankruptcy...heh...man did their timing suck)

later

I suppose. Just make sure they are clean, be gentle with the pins, and slow down a bit. The real nightmare is when you overheat the pins to the point where it breaks the connections inside the board. We can fix that.... :-)

Damn, litecoin is doing well these days. Amazing the value these Titans provide....

Another question I have is can I replace my PCIE plastic power connector by just plugging in a new piece over top the existing pins? I pulled off the plastic PCIE ports on 1 cube leaving the metal pins behind. I found some new PCIE plastic ports online that do not have the pins embedded into them. So can I just plug these new pieces on to the old pins and be back to BAU? I haven't received the PCIE male ports yet. But it took me a long time to find the part without the pins embedded into them. So I am hoping this works, without going through the whole solder process.

I finally found the time to replace the PCI-e connector on the cube where I cut one of the pins while trying to dismantle the plastic from the connector.
I didn't really expect it to take so much time, hell it's 6 pins. But all those that explained how much heat you need were perfectly right. What I found to be helpful is to have a hot air station + a soldering station (or a combined rework station). I did first heat the pin with the soldering iron so much I could apply some chipquik, after that I started to point also the hot air handle onto it and gave it 15-25s (more for the upper, longer pins) to get really hot.
Then I smoothly pulled with a plier on the pin until I felt they were completely lose and I could pull them out without resistance.

Spike53 was right: the shorter pins came out easier due not dissipating that much heat. I guess if you can't just throw ~100W heat at it, this is certainly a good idea to shorten them before trying to desolder.

While being at it, I applied new thermal grease (Gelid is great, but one syringe equals ~ one cube, so if you are planning to do more than one cube, get more than one syringe - I have to find a source where I get it in a little pot like in the video from Linus).

The warnings were justified, but if you have ever desoldered something and PCB and part were still usable, get a powerful iron and enough time and it will work, though as long as you can dismantle the plastics body, that is preferable. But when the connector is cut by an idiot like in my case or burnt beyond recognition you probably have no choice

What I would like to add as tip is to not use any force on pulling the pins out else you could risk to damage the via or pad.

Again thanks to all who helped with tips and links etc.!

I am having an unusual issue with my Titan, hopefully someone has an idea on how to proceed. 1 of my 4 cubes had the PCIE port fry and melt and it turned off. I removed this cube and ran on 3 cubes for awhile at a constant 240mhs with average heat. So I decided to reconnect the 4th cube and remove the PCIE plastic from the port and just connect directly to the pins that connect to the board. When I did this all 4 cubes got really hot within 3 minutes and dies started rebooting due to heat around 110C. I removed the 4th cube and the remaining 3 good cubes continued to get super hot and reboot.

So at this point I disconnected all cubes and booted up just 1 cube and it ran at 80mhs and heat was normal. But when I connected a 2nd cube both of them got super hot again. I then connected that 2nd cube by itself and it ran fine. Even the bad 4th cube ran fine by itself. For some reason any time I add more than 1 cube, all cubes get crazy hot. I tried using the reset button on the board and resetting it. I reflashed the card numerous times. I tried different data ports. Every cube works fine by itself on various data ports. But anytime more than 1 cube is connected they all get too hot to run. I am sooo confused, lol

yeah...i stalled to long on looking at my connectors. Today had a plug burn out on the Y adapter side. So thinking it was JUST the y adapter (the plug to the cube was cool to the touch
and not brown looking at it plug'd in) I tried to take it off to replace the y adapter thinking that was it. When I did get the y adapter out it had melted around the pins and now I have
to do your mod.

Then again looking inside it was dirty as heck....so will just leave it off and do a complete repaste...no time due to work...so probably it will be off for the week as i dork with the whole
repaste/maxumark brackets/heatsinks/fan and this mod project

Anyway now that there was NO HINT that anything was wrong on the cube end of the Y adapter. I suppose I will have to go thru each cube (before my cube by cube repaste) and look
to see if I have others working ok ..with a 'melded' connector waiting to go on the cube end.

Anyway just an FYI I guess I should have looked at both ends at all of this stuff monthly (I replaced all the knc y adapters with new ones like 2 months ago)....guess more then monthly
now that summer is here..just pull both ends look put back kinda schedule from now on

On a side note have got a deal on server psu's 2500w platitnum 95% eff (my corsair 1200i's are 91%) with backplate etc...and I think he said 15 gauge wire ..so has a DIRECT connect

to the PSU end and he uses NO Y adapter just driect connect to the cube and no issues (in that load wise they are setup for this kinda thing) so anyway for the price of a new corsair 1200i

I can get these new ..what the heck....probably the next way to go on the works (or at least the 6 cube unit) a bit of overkill

I found that cutting the pins short on the connector side, leaving enough that I could grab with forceps after hearing the I could grab with the forces and pull out pretty easily. The pins heat much easier when cut short. I used an iron with about 780 degree f.   

I pulled the connector off one of my cubes and replaced it.  Sorry for links instead of images, I haven't been around enough to post images.  Thought it might be worth sharing my experience.

Here is what my connector looked like:

http://imgur.com/IRFK9kw

Here is a shot of the old connector with the new connector next to it.  Notice I cut the pins before de-soldering.  This helped quite a bit because it allowed me to get in there and pull on the connector while de-soldering to get it out.

http://imgur.com/X9tl7Nw

This is the connector I purchased:

http://www.digikey.com/product-search/en?keywords=WM10869-ND

Like the folks here are saying, pulling the connector is not a pleasant experience.  I had only ever soldered once before after watching a video on youtube and replacing several capacitors on my television.  I thought I'd be brave and give it a go.  Though people said it would be hard I didn't really realize how hard until I gave it a go.  I picked up the quick chip and after watching youtube videos of that in use it appeared it would magically just pull right out.  Boy was I wrong.  The first weekend I tried I gave up after a couple of hours.  I decided to try again the following weekend and finally was able to get them.  I don't have a fancy soldering iron, just a $25 iron off Amazon.  The iron I have did have a heat setting that went all the way up to 450, I cranked it all the way up.  I think the thing for me on my first try was that I was not patient enough.  Because the pins are so thick, I think it takes a while to get them warmed up enough.  Maybe a pre-heat that people mention would have helped me there.  Anyways, as you can see in this picture, I was able to finally get them all pulled:

http://imgur.com/zF7SpvT

After replacing the connector I went ahead and made the modifications for cooling in this video as well as replacing all the cables on all 4 cubes.  I picked up a pair of 16awg cables off ebay.  Not sure if they're any good?  But, seem to be working so far.  The one gotcha on the cooling mods was the place on the case where the screws screw into to hold the board on came out.  I ended up resolving this with a bolt and a new screw.  As people mention, be careful not to apply too much pressure.

https://www.youtube.com/watch?v=-G_zcJ0vbOw

After the cooling mods my cubes run at ~10c cooler than they did before.  Just in time for summer.  All in all, I have no idea what I was doing and it may just be blind luck that my machine is still working.  Thought it may be helpful to share in case someone else gives it a go.  Good luck!

Hi all,
Has anyone found a source for the LCD display panel on the display module?  I have not been able to find a part number on this.

Thanks for any assistance.

So I did find the part.  Digikey
NHD-C0216CIZ-FSW-FBW-3V3-ND

Thank you Tupsu! I could order the coolers in the 9mm high version as the 14mm have no delivery date at all in my country. As I am not planning to run them on full clock speed, I hope these are good enough.



Thank you lightfoot, I found out meanwhile since I have opened two cubes that there is no second pad. I tried the trick to remove the plastics body and clean the pins as some of you here suggested. Worked great on the first cube, disaster with the second. The second connectors plastic was in better shape and didn't break that easy, and as the great craftsman with two left hands that I am I also have cut off ~2mm of the head of one of the GND pins... The following loud swearing should have been audible around the globe...
After all the warnings I received to avoid replacing the connector, I am now really friggin scared to do it.
I will have a good rework station, vacuum solder iron, chipquik alloy and replacement parts - are there any more tips that those that did succeed in replacing the connector would like to share to give me some confidence back *lol* ? I mean it is actually doable with enough power in the solder iron and alloy, right?

To be honest I have no plans to run the cubes over 250/275MHz ever. If they ever will work all at that speed, that will be a good miner to me and given all the time invested to get there I can surely resist any temptation to even try 300/325MHz even for a minute or two at all. I want to run it 24x7 and the experience especially with the pci-E connectors lets me seriously doubt if it's a good idea when running it at >=300MHz.

One note for those trying to dismantle the pci-E connector: Most often when you have a burned connector on the PCB, the socket on the y-connector is burned too and needs at least some pins replaced, often the connector housing too.

Thanks also for all the help I received by PM, I'll let you know how my Titan adventure turns out and eventually document the path.

Analyzing start-up log... sed: /var/log/initc.log: No such file or directory error code =
grep: /var/log/initc.log: No such file or directory Starting ASIC boards test...
InitA started in ASIC TEST MODE
BBB S/N: A335BNLT0A604013910873
LM75 OK. Temperature = 28.5°C
TPS65217 OK. Modification A, revision 1.1
Testing ASIC board #0... No board found
Testing ASIC board #1... No board found
Testing ASIC board #2... No board found
Testing ASIC board #3... error code = -12
Testing ASIC board #4... error code = -12
Testing ASIC board #5... error code = -12
Restarting mining...


What is this error?, Dies shows up,, cuve nr 4 and 5, hashes like normal.cube 3 dont hash.

what can it be?..............

Thanks for the insight lightfoot. It came this way from Sweden, it didn't develop this problem because it was run hard.

I remember when I first fired it up and noticed the problem. I talked to Kurt at KNC and told him I was more than a little disappointed that they had shipped me a defective Titan. He said "I don't think you understand what you purchased. We sold you 300 mhs scrypt, not a fully functional machine. If your machine hashes at a minimum of 300 mhs, then we've met our obligation' … I told him I completely understood what KNC had sold me, a piece of shit defective machine, delivered at the last possible moment, and I wanted a good one. At the time, KNC were ducking and dodging bullets from everywhere like crazy and I figured there was nothing to be gained by pursuing a complaint further. However I did succeed in eventually got banned from their forum.

All that said and done,  I'm kinda glad I have one now ;-) Thanks lightfoot for doing your part to keep them 'cooking'.

Are the 'high side FETs' internal to the Ericsson DC/DC module or external to them ?

Oh that's NEAT! What happened here is you lost a supply on die 4, most likely due to running it hard. However you didn't short the high side FETs, that's nice.

Now when you lose a die you can still run on one supply but only at half speed, say 150mhz for a 300mhz die. The die will hash at 150 happily and the supply will max out at around 40a. But you have it running at 225mhz which is impossible. Except that ohm's law is I=E/R, and if you boost the voltage you can run with less current. So running at 30% higher voltage allows you to run the die at a higher clock while pulling less current through the one remaining supply.

Darn, never thought of that. Weird as hell, but why not. Technically the limit is the Molex plug and since you're running one die slower (though still pulling a lot of power) you may be ok. I would not screw with it, run like hell... :-)

I have a Titan that seems to really like the heat. It runs very reliably 24/7 in ambient Texas summer heat of ~ 90F.
All the dies work at 325/-.0366 on all the cubes except one of the dies, which has to be down clocked to 225 and up volted to 0 to hash, and this die goes up and down every once in a while.  Its been like this since I received it from KNC.

The odd things about this cube with the marginal die is that one of the associated DC/DC converters is missing in the list on the advanced page. The display on the advanced page for this cube looks like this:



I'm reluctant to tinker with it as its been so reliable and especially because LTC mining looks set to become more profitable but I've also been kinda curious just what might be going on.

Any guesses ?

http://www.enzotechnology.com/where_to_buy.htm

http://www.aquatuning.de/luftkuehlung/passivkuehler/7045/enzotech-ram-kuehler-bmr-c1-high-profile-passiv

New Titan owner here - so please forgive if any of my questions are completely stupid.

I have two cubes with pretty coloured pci-E connectors (can't decide if this is still dark yellow or already brown  ) and I would like to replace them.

I have found the following part from Digikey and am wondering if that's the right connector or if there is a better one from DK or some other electronics warehouse:

WM1353-ND

Is it recommended to use some special solder for the pci-E connectors (i.e. lead-free with higher temperature) or can I take the my preferred day-to-day leaded solder?

Speaking of pci-E connectors, I have read rumours here that there is a second one on the PCB but not populated. Since I haven't opened any of the cubes I wonder if somebody can tell me if that second connector exists and simply can be populated and used to circumvent the single pci-E bottleneck?

Since I have to take the whole thing apart to replace the pci-E connector, I would also like to add the DCDC heatsinks from this video. Does anybody know where to get these?

As I have to re-apply the thermal grease and need to order that too: What thermal conductivity should it have at minimum? And any estimates on how many gram per cube I will need?

And finally: Is there a good source for 16AWG Y-splitters or do you guys make these all yourself? I realize the Mini-Fit connectors from Molex are only rated for 18-24AWG but I guess squeezing two 16AWG into such a crimp pin is just a question of enthusiasm - or are there better rated crimp pins than these:

WM2501CT-ND
WM2500CT-ND

Thanks for any tips, hints and links in advance.