GTX 1080 Founders Edition Died - 12V to Ground Short

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
Hey all, I'd appreciate some help troubleshooting this dead card - if we save it great, if not it is dead anyway...

My GTX 1080 FE died suddenly while gaming, I was overclocked and temps were reasonable ~mid 70C. The whole PC shut down, no lights on the motherboard etc.

TLDR: Trouble shot the system , it's 100% the graphics card, something has failed on it causing the 12V pins to short to the ground pins kicking the PSU protection off to save the system. My best SWAG guess is a mosfet, IC, or transistor is shorting?



I've seen others with GTX 1080 failures due to the tantalum capacitors but the failure methods seem different, but this poersons' issue sounds the most similar: https://www.reddit.com/r/overclocking/comments/cwjdte/gtx_1080_pcb_repairing/

This page has tons of good photos if mine below aren't good enough: https://xdevs.com/guide/pascal_oc/



I've stripped the card down, there is no signs of scorching, no smells of magic smoke, nadda that I can see.

I did a lot of research on graphics cards, their phase design, "stabbing" guides, and stalking RazorWind's different repair threads and videos, but I am certainly still learning a lot and have a lot to learn! I'm not 100% sure about all of the values I've found, and I'm sure I've made some obvious mistakes, so I'd appreciate any guidance on things to try (other than microwaving it :p), otherwise I may start with desoldering components to see when the short disappears (something I've seen Bob Kalpon and other youtubers do)...? I've also seen the Louis Rossman (IIRC) method of dousing the circuit board with alcohol to find hot spots (using an adjustable PSU) since I don't have an IR camera and want to avoid destructive testing, but for now I will poke around with my multimeter.

Card:
top.jpgbottom2.jpg

Power Phases:
phases.jpgphases 2.jpg





Probing Results (Resistance to Ground unless otherwise stated):
Using a milwaukee 2216-20 multimeter, with autosense resistance, it'll round down to 0.0Ω when there is very little resistance FYI.

For this on the mosfets I get these results, assuming the arrow is pointing at pin 1, and pin 4 should be the gate, and 5-8 are drains, I sussed this out since the datasheet doesn't specify for me https://pdf1.alldatasheet.com/datasheet-pdf/view/938419/MGCHIP/MDV3605.html

ina3221.jpg
q36
pin 4 3 2 1
~6M each
____________
pin 8 7 6 5
~4.2k each


q37
pin 4 3 2 1
0.7M, (10k going up?) ,13M , 13M - Need to clean this side up after likely butchering it with probe between pins 2&1 - measures could be off. I'll get my loupe out and see if it's gunk or a leg/solder later.
____________
pin 8 7 6 5
~6M each


q38
pin 4 3 2 1
0.4 0.4 0.4 118k
____________
pin 8 7 6 5
~6M each


The stripe on smd solid capacitors is the positive end, and on the electrolytes it's the negative, I added those measurements anyway. There was a lot of flux and some residue from the thermal pads so I had to dig in with the meter in some places.

From what I've researched, core resistance SHOULD be low, and the memory SHOULD be higher, I didn't find exact numbers for GTX 1080s though. I don't know about these measurements going up into the MΩ on C235,239,276,1176 either? I don't see any obvious shorts from these measurements..:

phases labeled 2eb.jpg


Older probe measures, with the mosfet pin measurements (the above picture supercedes any resistance measurements on the chokes and capacitors), there was flux or residue adding resistance to the ground measurements...
older probing w mosfets.jpg

Thanks for any and all suggestions!
 

RazorWind

Supreme [H]ardness
Joined
Feb 11, 2001
Messages
4,393
How confident are you in that 0.1 ohms to ground measurement on the 12V input connector? Is it super repeatable?

Edit: This was a great first post, by the way. I wish everyone who asked for help provided that much information.

Edit2: Where are you located, and do you have access to a hot air rework station and/or PCB preheater?
 
Last edited:

pendragon1

Extremely [H]
Joined
Oct 7, 2000
Messages
42,743
Edit: This was a great first post, by the way. I wish everyone who asked for help provided that much information.
its almost like hes seen some of your posts/help...
between this post and the other thread, you were first to come to mind. hope you dont mind.
edited speeling
 
Last edited:

pendragon1

Extremely [H]
Joined
Oct 7, 2000
Messages
42,743
I had another look at the OP's post this morning, and I think I have this figured out. We've now seen a bunch of failures like this here at [H] in my threads and in those posted by others. Can anyone who has been following along spot what's wrong here?
its missing a power phase at the bottom?! :)
 

Armenius

Extremely [H]
Joined
Jan 28, 2014
Messages
34,262
I had another look at the OP's post this morning, and I think I have this figured out. We've now seen a bunch of failures like this here at [H] in my threads and in those posted by others. Can anyone who has been following along spot what's wrong here?
The power sense resistor for the 12V on the top-right looks a little "bubbly." Also looks like there is a scorch mark on the PCB to the right of the caps near the bottom mounting hole.
its missing a power phase at the bottom?! :)
The empty pads on the power phase at the bottom is normal for the 1080 FE.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
How confident are you in that 0.1 ohms to ground measurement on the 12V input connector? Is it super repeatable?

Edit: This was a great first post, by the way. I wish everyone who asked for help provided that much information.

Edit2: Where are you located, and do you have access to a hot air rework station and/or PCB preheater?
Thank you! I try to be as thorough as possible.

The 12V to ground is very repeatable, I took the measurements thrice, and I did every pin each time. I would stick one probe into pin 1 and test 2-8, then pin 2, test 1 and 3-8, and so on. Any combo of ground and 12V gets me 0.1-0.4Ω every single time.

I'm in Oregon. I've worked with through-hole soldering a bit and done some basic smd repairs, but usually only things with 2-4 pads like capacitors, not ICs with many... No I don't have hot air nor a PCB heater, in the past I've ghetto rigged up a combo that worked but is finicky - a heat gun paired with an IR gun: warm everything up with quick passes and taking heat samples, and using a small butane pencil torch with this grid thing so only hot air comes out. It's a pain in the ass and I usually enlist a helper but I'm willing to get a rework station and a pcb heater, I've heard good things about Hakko but I'm open to any recommendations! Preferably in the hobbyist price range, but I'll invest in whatever I need.

I had another look at the OP's post this morning, and I think I have this figured out. We've now seen a bunch of failures like this here at [H] in my threads and in those posted by others. Can anyone who has been following along spot what's wrong here?
Beats me, that's why I'm here asking! Glad you've spotted something, hopefully it's something I can try to fix. The power supply cut power quickly so I'm hoping the GPU die/memory ICs didn't get bit with 12V...

its missing a power phase at the bottom?! :)
I wish it was that simple. You can see all of the power phases were basically cut in half to save money, plus cutting out a whole phase.
 
Last edited:

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
The power sense resistor for the 12V on the top-right looks a little "bubbly." Also looks like there is a scorch mark on the PCB to the right of the caps near the bottom mounting hole.

The empty pads on the power phase at the bottom is normal for the 1080 FE.

If you mean the shunt R53, the factory solder job looks like crap. I can't speak to the quality of the component itself.

I just tested resistance across the 3 shunts that I can see just in case and they all read 0.0Ω which is my meters' lazy way of saying milliohm I guess, which is about right for shunts since they affect the power readings - I've heard overclockers use resistors inline to cheat it for more power, but I haven't done that. Most of my measurements were otherwise from the ground pins of the 8-pin, unless noted (or forgotten to be noted...)

The shunts I checked: from the top of the card that is R52 at the bottom right, R51 at the middle of the right, and R53 at the top right by the 8-pin PCIe.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
I had another look at the OP's post this morning, and I think I have this figured out. We've now seen a bunch of failures like this here at [H] in my threads and in those posted by others. Can anyone who has been following along spot what's wrong here?
You don't think the GPU die is shorted/failed somehow... do you? I wasn't even pushing it hard I swear
 

RazorWind

Supreme [H]ardness
Joined
Feb 11, 2001
Messages
4,393
Beats me, that's why I'm here asking! Glad you've spotted something, hopefully it's something I can try to fix. The power supply cut power quickly so I'm hoping the GPU die/memory ICs didn't get bit with 12V...
If I'm reading your markings right, you have less than 1 ohm between the 12V input and the output of the core VRM to the GPU. What you should have there is a few million ohms, because there's a transistor in the way that switches the 12V on and off really fast to make the ~1.09V the core needs. You also have about 0.4 ohms to ground on all the 12V caps, which it should be noted is approximately the static resistance of the GPU core. What this strongly suggests is that you have the classic high side mosfet failure between the 12V input and the core power rail.

The good news is that there's a pretty good chance that the core is OK. I've never seen this on a 10 series card (because they're only just now becoming "old"), but I've had a number of 980 Tis pass across my workbench with this type of failure, every single one of them worked again after I cleared the short.

So, you have a couple of options for moving forward. One is to supply current at 1.0V into the 12V input and see if any of the components get warm, with concentration on the core power FET packages. Using some isopropanol as an indicator, if you don't have freeze spray or a thermal camera is a good way to improve this test. I would bet you strike out that way, though. All the components you're looking to test are made to handle huge current, and would hardly get warm even with 12V.

The other option is to start removing the FET packages until the short is cleared. Do not attempt this with a torch of any kind. You can use just a hot air station to remove them, but you really need a preheater to install new ones. You'll almost certainly destroy the new FETs when you install them without preheating the board. Those who have read my threads here at [H] may remember Solan's card, where I kept replacing the dead FET package, and it would keep burning out, and I'm pretty sure this was the reason. What I would probably do is remove the FETs from the core VRM one at a time until the short is clear, and then head the board back up, remove the remaining ones as well, and replace all of them with new ones. This includes the one for the memory rail, which I believe is the same part on this board.

If I were in the market for a hot air station, and I thought I'd use it very often, I'd probably try to use that as an excuse to buy the one that Louis Rossmann sells - the name of which escapes me now. Hakko is a good brand too, but pretty expensive. The one you see me use on Youtube is a cheap Aoyue 968A, which includes a pretty nice soldering iron. It's cheap, but effective. I also have the INT866 combination preheater/soldering station, which I like way more than I thought I would. If you were going to fix a lot of cards, you'd want something more powerful, though, like the fancy Hakko one.

If you're ordering a hot air station, make sure the one you get is meant for 110V power. 220V versions exist as well, and won't work in the US unless you have a monster dryer circuit to plug them into.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
Thank you so much for your input!

I will see if my 12V meanwell PSU adjusts that low, I don't think it does, so onto fet removal. I agree the torch method would be too reckless for components like that, I feel like a shunt or simpler circuit I could skate by, but mosfets I'd probably rip a trace or burn something out. This is the torch I was talking about, some of the interchangeable bits let you do hot air only (the flame is expelled out of a port on the side), so I'm not a complete lunatic! As far as I recall there is only one setting to have the flame come out for working as a proper torch, the rest are for soldering/hot air: https://www.lowes.com/pd/BernzOmatic-Lead-Free-Soldering-Kit/1000170991 , it is good for when you need to repair something appropriate for a typical iron and don't have power - I can do heat shrink and solder, and this model doesn't need a lighter, but SMD repair is already tough enough so I will get the appropriate tool.

The fet rework will definitely take more finesse than what I was talking about, so I think I will invest in the Aoyue 866. I like that it's an all in one at a decent price, unless you think I would do better else-wise.

Louis Rossman sells the Atten ST-862D for about $250, but I honestly think it'll be overkill for my usage, and I'd still want the preheater. If I have the kit I will use it, but certainly not often enough to expect it to die for several years. Otherwise I'm looking at $400+ for hot air and preheat. I'll also need a new flux paste syringe, I should still have wick, is thin gauge 60-40 solder is fine or is there something better? I will experiment with a dead 9800 GTX+ to get a feel for SMD reworking before trying on the 1080. Anything else I'm missing? I don't have a silicone mat anymore either, probably won't need one with the Aoyue 866 anyway?



I'll probably get 10 mosfets, NTMFD4C85N:

Ebay ~$20, but will it be any more reliable than alibaba in terms of quality?
https://www.ebay.com/itm/5pcs-NTMFD4C85NT1G-NTMFD4C85N-4C85N-QFN8-new-/123248646656?_ul=IL

amazon ~$34, same issue as above. Still coming from China
https://www.amazon.com/4C85N-NTMFD4C85N-NTMFD4C85NT1G-Chipset-Original/dp/B0838TSTTV

Mouser ~48$
https://www.mouser.com/ProductDetail/ON-Semiconductor/NTMFD4C85NT1G?qs=dbuNSGnowt0IPuJj8dK4RA==

Digikey says the have them, I can't seem to add them to my card, not sure if they have a bug or what but looks like I'd have to order 117 of them?
https://www.digikey.com/en/products/detail/on-semiconductor/NTMFD4C85NT1G/11529722

If you think it's all the same I'll snag up some ebay fets, but I'll happily opt for whatever is recommended.




I know card repair is partially experimental, how do you feel about me adding mosfets where NVIDIA held back, restoring it to two per phase? In my mind that would cut the workload/stress in half, hopefully preventing future failures. I checked the pads just in case, they are all connected, so from my understanding there is no reason I couldn't or shouldn't?

I could probably re-add the missing phase too but that TI 53603A and three resistors and capacitor look like a royal PITA being so small, and that's assuming I could reasonably source them and something else isn't missing or changed elsewhere on the PCB so I'll likely can that idea as a fun thought experiment. The card did survive for probably four years of mild gaming with the 5 half-neutered phases.


The memory does use the same mosfets so I'll order a bunch, it looks like I forgot to list the major components, which I'll do so now for posterity's sake:

GPU:
Core: Nvidia Gp104-400-A1 Core
Memory: Micron D9TXS; MT58K256M32JA-100:A 8Gb: x16, x32 GDDR5X SGRAM with rated data rate at 10.0 Gbps

Power Phase:
TI 53603A Dual phase step down controller? Not 100% sure, it's the 8-pin model. Finding the datasheet for the 8-pin is rough.
ON NTMFD4C85N Dual PowerPhase FET , powering the 5 phases to gpu and the 1 to mem
Tantalum ( I believe) Capacitors 330µF and 470µF
R22 and R33 chokes, not sure on the ratings, 12:1 on the r22 maybe?
uP9511P 8/7/6/5/4/3/2/1-Phase Synchronous-Rectified Buck Power Controller
Texas Instruments INA3221 Power Sensor IC for current monitoring, works with the shunts AFAIK

There is probably more but I'll have to double check, I'll edit them in later if I find any major players.
 
Last edited:

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
If I'm reading your markings right, you have less than 1 ohm between the 12V input and the output of the core VRM to the GPU. What you should have there is a few million ohms, because there's a transistor in the way that switches the 12V on and off really fast to make the ~1.09V the core needs ... What this strongly suggests is that you have the classic high side mosfet failure between the 12V input and the core power rail.
Just to make sure I understand correctly:
The output of the core VRM to the GPU measurements are the coils/chokes which are supposed to step down to the core voltage?

And the transistors you speak of that flop are the mosfets? Just making sure I'm on the same page over here.

I thought it was quite odd that the memory mosfet read in the MΩ range, but I wasn't sure what number to expect either. As far as I understand now, the mosfets are all working in parallel, so that is why I would have to desolder them all to find the bad culprit?

And going off of that assumption, that is why I'm wondering if I can do an upgrade by adding mosfets to the empty pads since they are all electrically connected anyway (skipping the missing phase)?

Looking at the card again, it looks like it originally was designed to be 6 core phases, and two memory phases with a shared choke? I'm not sure though since there is an odd number of mosfet pads in that location, but 16 in total.
 

Nobu

[H]F Junkie
Joined
Jun 7, 2007
Messages
8,649
Just to make sure I understand correctly:
The output of the core VRM to the GPU measurements are the coils/chokes which are supposed to step down to the core voltage?

And the transistors you speak of that flop are the mosfets? Just making sure I'm on the same page over here.

I thought it was quite odd that the memory mosfet read in the MΩ range, but I wasn't sure what number to expect either. As far as I understand now, the mosfets are all working in parallel, so that is why I would have to desolder them all to find the bad culprit?

And going off of that assumption, that is why I'm wondering if I can do an upgrade by adding mosfets to the empty pads since they are all electrically connected anyway (skipping the missing phase)?

Looking at the card again, it looks like it originally was designed to be 6 core phases, and two memory phases with a shared choke? I'm not sure though since there is an odd number of mosfet pads in that location, but 16 in total.
You'd have to clear the short, and add all the other missing components for that phase. Not sure if the firmware would control that phase either, may have been nuked?
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
You'd have to clear the short, and add all the other missing components for that phase. Not sure if the firmware would control that phase either
That would be my fear as well, I'm much more interested in doubling up the mosfets on the existing 5 phases than reintroducing the missing 6th.

I've researched down some rabbit holes, sometimes a simple resistor swap is enough to pull a controller to different phase counts, but I don't know really. In the end it wouldn't be worth it especially if I could double up the mosfets on the 5, since mosfets are what likely failed anyway.
 

RazorWind

Supreme [H]ardness
Joined
Feb 11, 2001
Messages
4,393
Just to make sure I understand correctly:
The output of the core VRM to the GPU measurements are the coils/chokes which are supposed to step down to the core voltage?
Yes. The primary function of the inductors is to smooth the alternating 12V/0V out, producing a sawtooth-ish approximation of 1.09V. The strip of copper between them and the positive side of the capacitors is wired directly to the GPU core.

And the transistors you speak of that flop are the mosfets? Just making sure I'm on the same page over here.
Yes. The flat black components to the right of the chokes are dual N-channel mosfet packages. There are two transistors inside each one - a high side and a low side, sometimes called "upper" and "lower". The high side one switches the 12V input power on and off. The low side one switches the flyback current from the inductor on while the high side is off, so that it powers the GPU as well. On your card, one of the high sides has failed, creating a situation where the 12V is on all the time from one of the phases.

If you look at the datasheet for the 4C85N, you'll see that the drain of the low side and the source of the high side are wired together to one of the exterior terminals, which is called the switch node. The switch node is the terminal that is connected to the chokes, and from there to whatever component you're powering.

I thought it was quite odd that the memory mosfet read in the MΩ range, but I wasn't sure what number to expect either. As far as I understand now, the mosfets are all working in parallel, so that is why I would have to desolder them all to find the bad culprit?
It depends on where exactly you're measuring. If measure from the switch node to ground, you're actually measuring the resistance through the memory circuit, which should be in the tens of ohms on this card (30-40, maybe?). If you measure from the 12V input to ground, you'll get a much higher number. You should get a lower, but still high, number if you measure at either gate pin.
And going off of that assumption, that is why I'm wondering if I can do an upgrade by adding mosfets to the empty pads since they are all electrically connected anyway (skipping the missing phase)?

Looking at the card again, it looks like it originally was designed to be 6 core phases, and two memory phases with a shared choke? I'm not sure though since there is an odd number of mosfet pads in that location, but 16 in total.
I wouldn't attempt to populate the missing pads, but you may be able to use them if you damage the main ones. Concentrate on just getting the board working first, and then worry about modifying it, if you feel so inclined.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
Thanks RazorWind, That all makes a lot of sense and answers questions I didn't yet ask even!

Sounds good, and good idea - try to fix it first in case there are any other issues. Any opinion on component sourcing, or just get it however I can?
 

RazorWind

Supreme [H]ardness
Joined
Feb 11, 2001
Messages
4,393
Thanks RazorWind, That all makes a lot of sense and answers questions I didn't yet ask even!

Sounds good, and good idea - try to fix it first in case there are any other issues. Any opinion on component sourcing, or just get it however I can?
I've heard tell of some folks getting counterfeit parts from ebay, but I've never actually had that happen myself. I say go for it.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
I've heard tell of some folks getting counterfeit parts from ebay, but I've never actually had that happen myself. I say go for it.
Order says ~June 17th, it's on the slow boat but that gives me a chance to practice on some old junk cards!

I have one with a bad PCB, you can hear crunching when lightly flexing it, I'm assuming traces. Found that out pulling the heat sink, it's a low end old card anyway, or I have the 9800 gtx+. I can play doctor and pull, retin, and reattach some mosfets on those to get the hang of it.
 

WilyKit

Limp Gawd
Joined
Dec 18, 2020
Messages
457
One of the most entertaining and informative threads on [H] I’m anxiously awaiting the results of the repair.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
First of the tools came in! I got a cheap USB microscope to help make sure I don't miss anything! I also got more solder wick, flux, etc but the reflow station won't be in until Monday or later.

Below here are some closer images of the main components needing addressed. I'm not sure if Q37 was smooshed' when I was probing, regardless that is a solder blob by the look of it. I should have used safety pins instead of the probes for testing resistance, oh well it's too late now. It's also blocked in part by that cap, so I'm thinking Kapton tape and tinfoil to isolate it from neighboring components, that might help keep me from blowing other components away? Kapton is good till about 500ºF supposedly, I've had luck with it and tinfoil in the past.

Otherwise I can rig up a very fine soldering tip if the kit I got doesn't have one fine enough, although if I could jab a probe in there I'm sure I could fit a very fine tip (or perhaps I'll wrap some copper wire around the tip for a DIY thinner tip) ... I'll test the Kapton on a sacrificial card first to see how well it protects the little components from getting blown away to sate my curiosity, but I'm liking the thinner tip idea for Q37 because of the close proximities...

Q37 MDV3605 MOSFET https://datasheetspdf.com/pdf/1107906/MagnaChip/MDV3605/1 (I believe):
v3605 pin smoosh.jpg

Core and Mem ON NTMFD4C85N Dual PowerPhase FET (4C85N MOSFET) up close:
4c85n mosfet.jpg4c85n mosfet2.jpg

That's all for now. The new MOSFETs won't be here till mid-June'ish, but I can start pulling them one by one to see if/when the 12V/ground short drops off at least when the hot air kit comes in.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
I've got everything but the mosfets and air flow setup, so I whipped out the test victim. I don't want to go in cold on my GTX 1080 so I will be practicing on a 9800 GTX+ that died, I wish it worked for one of my old socket 939 builds but oh well.

Being antsy and wanting to experiment I tried the heat gun "preheat", with the micro torch on hot air. That setup was actually very serviceable, although you would want a helper to be able to pass off the heat gun back and forth as needed to maintain warmth - I used it to get the area evenly warmed, then using the butane torch hot air attachment for the actual desoldering. It only passes hot air, the flame is vented out the sides, and it just gets hot enough to melt the lead-free solder.

I used an old scrap of 3/4 ply underneath, preheating that as well with the heat gun so it would draw less heat from the PCB, and protect the work surface. RazorWind don't worry, I wouldn't dare try this on the GTX 1080! The hot air / preheater combo is coming, I'm just impatient to start practicing and getting a feel for SMD work again.

Sadly my torch ran out of butane right after removing the uP7706AU8 IC, the closest in size and having a similar layout to the mosfets on the GTX 1080, albeit with longer legs.

Do NOT try to do any reflow work with just a heat gun, they blow too much air which will cause the smaller components to float away. I may have tested a section just to see how it would do, it was BAD!

Anyway, pics:

Before
the bigger victim.jpgthe victim.jpg

After pulling the IC, running out of Butane on reassembly, murdering everything with the heatgun:
the IC by itself.jpgoh the humanity.jpg

I cleaned up the pads after the heatgun incident, before I got tired of bothering. You can see C37 halfway falling off still. I did remount the IC ultimately, and get all but one component tacked back on with my soldering iron, but a hot air reflow (or my butane) would do it much better. One of the tiny resistors floated away or stuck to something, oh well.
cleaning up after hg.jpg (Forgot after pic, may add later)
 

RazorWind

Supreme [H]ardness
Joined
Feb 11, 2001
Messages
4,393
I've got everything but the mosfets and air flow setup, so I whipped out the test victim. I don't want to go in cold on my GTX 1080 so I will be practicing on a 9800 GTX+ that died, I wish it worked for one of my old socket 939 builds but oh well.

Being antsy and wanting to experiment I tried the heat gun "preheat", with the micro torch on hot air. That setup was actually very serviceable, although you would want a helper to be able to pass off the heat gun back and forth as needed to maintain warmth - I used it to get the area evenly warmed, then using the butane torch hot air attachment for the actual desoldering. It only passes hot air, the flame is vented out the sides, and it just gets hot enough to melt the lead-free solder.

I used an old scrap of 3/4 ply underneath, preheating that as well with the heat gun so it would draw less heat from the PCB, and protect the work surface. RazorWind don't worry, I wouldn't dare try this on the GTX 1080! The hot air / preheater combo is coming, I'm just impatient to start practicing and getting a feel for SMD work again.

Sadly my torch ran out of butane right after removing the uP7706AU8 IC, the closest in size and having a similar layout to the mosfets on the GTX 1080, albeit with longer legs.

Do NOT try to do any reflow work with just a heat gun, they blow too much air which will cause the smaller components to float away. I may have tested a section just to see how it would do, it was BAD!

Anyway, pics:

Before
View attachment 358495View attachment 358497

After pulling the IC, running out of Butane on reassembly, murdering everything with the heatgun:
View attachment 358496View attachment 358494

I cleaned up the pads after the heatgun incident, before I got tired of bothering. You can see C37 halfway falling off still. I did remount the IC ultimately, and get all but one component tacked back on with my soldering iron, but a hot air reflow (or my butane) would do it much better. One of the tiny resistors floated away or stuck to something, oh well.
View attachment 358493 (Forgot after pic, may add later)
Didn’t you say you ordered an Aoyue 866? Why not use that? Also, you need flux.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
Didn’t you say you ordered an Aoyue 866? Why not use that? Also, you need flux.
I got flux and the Aoyue 866 will be here soon, it's delayed now to May 27th... just a practice run in futility, that card has a broken PCB so it's just good for practice unless anyone needs a component. There's a couple other ICs that are mosfet-like that I am saving for the Aoyue practice runs.

The flux is Chip Quik NC191, I wanted to make sure that stuff worked OK, since I almost opted for the MG Chemicals flux at twice the price. My 60/40 solder worked fine to retin the components but I'm thinking I'll need to get fluxless solder, not this rosin core?

Thanks again for all the help
 

KD5ZXG

Gawd
Joined
Mar 24, 2017
Messages
704
If you mean the shunt R53, the factory solder job looks like crap. I can't speak to the quality of the component itself.

I just tested resistance across the 3 shunts that I can see just in case and they all read 0.0Ω which is my meters' lazy way of saying milliohm I guess, which is about right for shunts since they affect the power readings - I've heard overclockers use resistors inline to cheat it for more power, but I haven't done that. Most of my measurements were otherwise from the ground pins of the 8-pin, unless noted (or forgotten to be noted...)

The shunts I checked: from the top of the card that is R52 at the bottom right, R51 at the middle of the right, and R53 at the top right by the 8-pin PCIe.
Sounds like you need a 4wire ohmmeter, or fake it with two meters.

One on Ohms only to send the test current. Don't use to measure. Always slightly off due to its own cable resistance.

The other on Volts to measure the drop across devices in question. Without the distraction of send current in the cables.

Might also want to solder send and sense together near the tips, so you are only handling one pair of probes, not four.

Some call this a Kelvin connection, not entirely sure why.

If 1st meter on autoscaled ohms don't present a constant current, could also try diode check for the send current.

Receive volts may not tell you actual ohms, but it also won't be measuring a zero or jumping around.

Then go hunting for the shortest short. Should be able to tell which half of the board anyway.

 
Last edited:

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
The hot air reflow station arrived today, so I'll do some practice runs properly this time around on the test victim 9800 GTX+ first. If that goes well, I'll pull mosfets down and see if the short disappears.

KD5ZXG thanks for that information! I've heard about 4-wire measurements but hadn't considered it. I'll keep learning about that.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
Mosfets show as having arrived. We'll see how far I get this weekend. My plan is to begin pulling the old mosfets with my new hot air setup, after a little more practice on the sacrificial 9800 gtx+
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
RazorWind Thanks again for your contributions both to me and the community. I've learned a lot from you.

I finally set some time aside to give this another go. I got the Aoyue 866 and have been testing it with my old sacrificial card, getting comfortable with the settings since last I updated. I wanted to comfortably be able to remove and reattach SMD components before making the attempt..:

Well I finally got the nerves to try it on the GTX 1080! I decided to remove all five GPU core MOSFETs versus trying to figure out which one(s) were bad. After pulling all 5 the 12V short to ground is GONE. Resistance between 12V and ground is in the kΩ range now, inline with what was said before for a working card.

Testing the MOSFETs one by one, I found a single one that has a short between source 1 and drain 1. I tested it both ways because these is supposed to be a diode between them. I'm not the best at reading circuit diagrams but from what I can figure out is this Dual N-Channel MOSFET design shares drain 1, with source 1 and 2 and gate 1 and 2 driving both halves to it? I'm not sure what SW 5 6 7 do though, but from my research having virtually no resistance between a source and drain is BAD :). I'd love a simple explanation on the attached datasheet in case I'm off base. From my research on MOSFET details and trying to read the schematic: the source pins 1 & 2 should be negative, with the gates determining flow to the drain from the sources? What part do the SW 5/6/7 pins have, is that short for switching? The schematic makes it look like the lower MOSFET inside the package feeds the second?

The bad news: I was sent the wrong MOSFETs. My bad for not checking sooner. I am tempted to resolder the "good" four and testing 2D at least if the 12V short stays gone. Not using it, just making sure the card isn't dead, I've read about some others who were able to test the card could at least post momentarily on a single phase, so 4 outta 5 should be good enough?
 

Attachments

  • 2021-08-03 22_10_16-NTMFD4C85N - PowerPhase, Dual N-Channel SO8FL - NTMFD4C85N-D.PDF — Mozilla...png
    2021-08-03 22_10_16-NTMFD4C85N - PowerPhase, Dual N-Channel SO8FL - NTMFD4C85N-D.PDF — Mozilla...png
    118.6 KB · Views: 0
  • mosfets compared.jpg
    mosfets compared.jpg
    253.9 KB · Views: 0
  • mosfets out.jpg
    mosfets out.jpg
    390.3 KB · Views: 0

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
Got 2 out of the 5 original reattached. I was too excited when removing the first two that I broke one of the tiny single pads off each. I'll be more patient in the future as I was on the last two, which floated off before I interfered.

Resistance is still good between 12V and ground, so I will probably try a short boot up and see if it posts to BIOS. With the short gone the PSU won't kick us off so now it's worth finding if the card is dead or not.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
AAAAAAAAAAAAND IT WORKS!!!!!!!

I setup a very temporary test bed, put the four main screws that clamp the cooler to the die on and skipped the rest, connected the fan and LED strip, and gave it a try. Booted up perfectly fine, posted BIOS screen and everything! I don't want to stress it with 3D loads yet with only two MOSFETs running the core, if I'm patient enough I may try to remove some of the casing on the two that I broke and see if I can get at the wires, but ultimately I've ordered another five MOSFETs that actually show the pin out on the bottom and it matches these, so another two weeks we can really give it a go!

Thanks again RazorWind , so so much!
 
Last edited:

RazorWind

Supreme [H]ardness
Joined
Feb 11, 2001
Messages
4,393
AAAAAAAAAAAAND IT WORKS!!!!!!!

I setup a very temporary test bed, put the four main screws that clamp the cooler to the die on and skipped the rest, connected the fan and LED strip, and gave it a try. Booted up perfectly fine, posted BIOS screen and everything! I don't want to stress it with 3D loads yet with only two MOSFETs running the core, if I'm patient enough I may try to remove some of the casing on the two that I broke and see if I can get at the wires, but ultimately I've ordered another five MOSFETs that actually show the pin out on the bottom and it matches these, so another two weeks we can really give it a go!

Thanks again RazorWind , so so much!
Right on!

I would have told you not to try reusing the old FETs. Heat is the enemy of any semiconductor, and you've now heated those poor things way beyond their operating range at least twice, and they're designed to survive being soldered only once. I can almost promise that if you try running anything more strenuous than the BIOS UI on it, you'll burn out another one of the high side FETs. The core usually survives taking 12V on the chin a few times, but especially these days, it's better to err on the side of caution.

If you intend to actually use this card, get new ones for all five core phases and replace the old ones. These are super easy to solder, so it should be a piece of cake with the 866. You might want to consider replacing the one for the memory as well. You've heated that one up at least twice as well, even if you didn't remove it.

Regarding your question about how the dual N-FET works, maybe thd diagram below will help. Click to embiggen. The diagram in the datasheet doesn't mark the switch node very well, which you can think of as the output to the logic IC.

Every FET has three terminals:

The source - this is connected to the low or negative voltage, usually ground. It's called the source because electrons flow into the transistor through this terminal. Remember that electrons have a negative charge.
The drain - this is connected to the high voltage. It's called the drain because the electrons flow out through this terminal
The gate - When a sufficiently high voltage is applied to this terminal, the transistor turns "on" and the resistance between the source and the drain drops to almost zero, allowing current to flow through the transistor, almost as if it were a wire. When the gate voltage goes away, the transistor turns off and stops the current flow.

A dual N-FET package, such as the 4C85N has two transistors inside of it - the high side and the low side, with the drain of the low side and the source of the high side wired up together and to a terminal on the outside of the package.

IMG_0489.JPG
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
I'll leave it off until then, I believe i got a 6 or 10 pack of fets so sounds like a plan, I just wanted to make sure it works before putting in the extra work.

Thanks again!
 

Falkentyne

[H]ard|Gawd
Joined
Jul 19, 2000
Messages
1,843
Right on!

I would have told you not to try reusing the old FETs. Heat is the enemy of any semiconductor, and you've now heated those poor things way beyond their operating range at least twice, and they're designed to survive being soldered only once. I can almost promise that if you try running anything more strenuous than the BIOS UI on it, you'll burn out another one of the high side FETs. The core usually survives taking 12V on the chin a few times, but especially these days, it's better to err on the side of caution.

If you intend to actually use this card, get new ones for all five core phases and replace the old ones. These are super easy to solder, so it should be a piece of cake with the 866. You might want to consider replacing the one for the memory as well. You've heated that one up at least twice as well, even if you didn't remove it.

Regarding your question about how the dual N-FET works, maybe thd diagram below will help. Click to embiggen. The diagram in the datasheet doesn't mark the switch node very well, which you can think of as the output to the logic IC.

Every FET has three terminals:

The source - this is connected to the low or negative voltage, usually ground. It's called the source because electrons flow into the transistor through this terminal. Remember that electrons have a negative charge.
The drain - this is connected to the high voltage. It's called the drain because the electrons flow out through this terminal
The gate - When a sufficiently high voltage is applied to this terminal, the transistor turns "on" and the resistance between the source and the drain drops to almost zero, allowing current to flow through the transistor, almost as if it were a wire. When the gate voltage goes away, the transistor turns off and stops the current flow.

A dual N-FET package, such as the 4C85N has two transistors inside of it - the high side and the low side, with the drain of the low side and the source of the high side wired up together and to a terminal on the outside of the package.

View attachment 381444

This is the first good news I've read in this entire forum in many months. Someone resurrecting a dead card without a fried core. Very interesting.
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
That chart and explanation helped greatly! What does iw mean next to the 12V on the top D1 section mean?

When I look at their rendering on the left of the datasheet combined with your information my head hurts further: how is work being done if the two mosfets are joined source to drain, with that joining section also going to do the work of powering the IC with Vout?

A single mosfet I can wrap my brain around, but these dual mosfets throw me. I assumed electrons on the low side are flowing from the source to the switch node which would've been the drain on a single mosfet, and had assumed that before based on the direction of the diode, but that logic doesn't apply to the high side with the diode oriented the same way - my logic would have that being a P-channel mosfet too I guess. I'm missing something key here and I'm sure my ignorance is showing.

I thought with this being a dual package mosfet that electrons would be going through the high side drain to what would have been source too at the switch node, ultimately to the IC before going to a positive plane? In the diagrams the diode on the high side would be allowing electrons out of the drain? Apparently the diode is normal for mosfets and is a parasitic component but again I'm not understanding the electron flow when considering the high side. Are these mosfets being switched on and off for PWM, with the high and low firing out of sequence to fill the PWM gaps (or doubling voltage or adding their amperage)?

Another forum on high vs low mosfets had this to say:
""Low-side" means the current travels from the load or device through the mosfet to ground (common). "High-side" means the current travels from the supply through the mosfet to the load and then to ground.

Another way to put that is:
Low-side = mosfet source to ground, drain to load, load to supply.
High-side = mosfet drain to supply, source to load, load to ground."
Source: https://forum.allaboutcircuits.com/...sfet-high-side-and-low-side-switching.124664/

With them being joined and that descriptor above and the labeling, it sounds more like AC going back and forth... which I could imagine if you toggle each gate on and off oppositely if the dual-fet package has both negative and positive connections and the receiving end was able to reciprocate... but ICs are DC powered. Obviously I'm on the crazy train now.

Is there any logical reason outside of manufacturing preferences that pins 5/6/7 aren't joined like D1?
 

rhkcommander959

Limp Gawd
Joined
Mar 26, 2020
Messages
334
Sorry for the long delay, I had to reorder the mosfets after receiving the wrong models, found some that should do the job and have finally installed them!

Last I left off the GPU worked with one (test) mosfet. I've now replaced them all and am waiting for the card to cool down before I bolt it back together! The last core MOSFET was giving me trouble, I think I had just a touch too much solder on it, I clamped onto the ground and 12V pins to monitor resistance and it suddenly was shorted on the last mosfet. popped it off, tested it direct, saw a tiny sliver of solder bridging it, retinned, used copper braid to remove excess, and popped it back on. Try 3 it was good to go so the second this card cools down I will give it a rough test assembly, see if it posts, and then full assembly if it does! I'll post back if the card works, fingers crossed!

Thanks again RazorWind , I never would have taken the leap without your guidance!
 
Top