Dead Graphics card. 780TI. Dead Mosfet perhaps?

[Sikator]

Hey. First of all i am a brand new member here, and i hope i am doing things right here in my first post.

 I've had a Dead GTX 780TI for quite a while now. i had a post on a completely different forum where we ended up with me having to test the mosfets and then the person helping me dissapared after saying so.
So here i am, months later hoping someone could help me figure out how i am supposed to test these mosfets.

I simply googled "Electronics expert forum" And here i am.



Info:
Name: 780ti Palit Jetstream
Model: NE5X78TH10FB-1100J

Symptoms:
So to make it simply.
I added 200% on the power limit and clicked apply. (Possible thanks to a Custom GPU bios)
Fan speed went to 100%
1 of 3 screens went black and the one with MSI Afterburner (Overclocking software) froze to let me see my terrible mistake for a couple seconds.
and the pc shut down after that.

As i understand, just doing this shouldn't have killed it, though that was the last seconds of its life.


I managed to flash the bios back to my backup by unplugging the power cables to the GPU itself and turning on the pc using a second GPU to get image.
Even tho it was a successful flash, if i have the power in before i start up my pc the fans are instantly set to 100% and the 780ti is not recognized. So nothing changed.








The section:

The mosfets:

And the things next to them:

This is the mosfet right if i am not mistaken http://www.vishay.com/docs/63788/sic780.pdf



Would be really nice to know if these were dead or if something else is the reason of the card's death and a possible fix.

Thanks in advance!




EDIT: If i turn on the pc with the GPU and Without Internal power connected (PCI 8x2) It will print a message on the screen "Please power down and connect the pcie power cable(s) for this graphics card"

[Armadillo]

Use a multimeter ohm range, probe across the 2 pads of the 8 Sic780s, which of them are the lowest of ohms?

[Rasz]

>EDIT: If i turn on the pc with the GPU and Without Internal power connected (PCI 8x2) It will print a message on the screen "Please power down and connect the pcie power cable(s) for this graphics card"

External power, internal is thru PCIE. This is a good sign!

Measure between pins landing on a pad with U12 printed on(lower left) and the ones marked red in lower right corner of the chip.

I suspect one of the mosfets is shorted between Vin and ground, and luckily its the one powered from external socket = you can still boot without external power and get usable picture with "please plug blabla" message.

[Sikator]

Use a multimeter ohm range, probe across the 2 pads of the 8 Sic780s, which of them are the lowest of ohms?

With my multimeter set at (200 ohm reading) I get 5.1-5.2 on all the chips no matter what configuration of From pin-To pin i choose.

[Armadillo]

Sic780 seems OK.

With the multimeter ohm range, all the 4901NF you can find [power mosfet], measure the pads as shown, which ones have the lowest of reading?

[Rasz]

4901NF

seems to be ram supply rails, and they have to work for the GPU "please plug external blabla" message to appear.

Edit: 3rd Edit: if you look closely at the pcb 4 bottom Sic780s are supplied from PCIE connector - those are fine
next 3 from the right external power, upper Sic780 and bottom 4901NF is supplied by left external supply

https://www.ekwb.com/configurator/upload/pictures/Palit-GeForce-GTX-780-Ti-JetStream-3GB-GDDR5-(NE5X78TH10FBJ)-PCB_54863.jpg

http://gecid.com/data/video/201404241300-5413/img/08_palit_ne5x780h10fb-1100j.jpg


so bottom 4901NF, or one of 4 upper Sic780 has to be broken
....oooooor external power sensing is broken. This card requires BOTH external connectors to be plugged, otherwise it will act just as described (all fans full tilt + no boot).

btw mosfets can be fried full open too+ controllers also die, so if you cant find a short you will have to measure on powered card and look at the gates and outputs of every single transistor (with a scope)

[Sikator]

Sic780 seems OK.

With the multimeter ohm range, all the 4901NF you can find [power mosfet], measure the pads as shown, which ones have the lowest of reading?

Well there are two of these chips. And they mesure different.
But i realized some of them mesure different from where i put my black and red probes. some are

These are the Top power mosfet




These are the Bottom power mosfet



I am trying to mesure from the two other pictures you posted but they are up and down everywhere o.O

[Sikator]

>EDIT: If i turn on the pc with the GPU and Without Internal power connected (PCI 8x2) It will print a message on the screen "Please power down and connect the pcie power cable(s) for this graphics card"

External power, internal is thru PCIE. This is a good sign!

Measure between pins landing on a pad with U12 printed on(lower left) and the ones marked red in lower right corner of the chip.

I suspect one of the mosfets is shorted between Vin and ground, and luckily its the one powered from external socket = you can still boot without external power and get usable picture with "please plug blabla" message.

4901NF

seems to be ram supply rails, and they have to work for the GPU "please plug external blabla" message to appear.

Edit: 3rd Edit: if you look closely at the pcb 4 bottom Sic780s are supplied from PCIE connector - those are fine
next 3 from the right external power, upper Sic780 and bottom 4901NF is supplied by left external supply

https://www.ekwb.com/configurator/upload/pictures/Palit-GeForce-GTX-780-Ti-JetStream-3GB-GDDR5-(NE5X78TH10FBJ)-PCB_54863.jpg

http://gecid.com/data/video/201404241300-5413/img/08_palit_ne5x780h10fb-1100j.jpg


so bottom 4901NF, or one of 4 upper Sic780 has to be broken
....oooooor external power sensing is broken. This card requires BOTH external connectors to be plugged, otherwise it will act just as described (all fans full tilt + no boot).

btw mosfets can be fried full open too+ controllers also die, so if you cant find a short you will have to measure on powered card and look at the gates and outputs of every single transistor (with a scope)

Interesting, first, you ment like this right?

and i would test like this on all four top SIC780's?


EDIT: If so, Most gives 3.3 or 3.0 measured with 20k on my multimeter. Some configurations gave 0.00 but so did all the others in the same configurations.
Now the 4901NF confuses me a bit aswell. the two 4901NF's do act different in mesurements as showed futher up.

I  am sorry if some of the info i am giving is wrong or weird. I really am not good at this sort of stuff but i will be able to get someone to change parts like this if needed without paying a ton. Which they wanted if they had to test stuff. idk why.

[Armadillo]

Emm!... don't seems like dead short to me.... before we conclude the GPU,......
the external power, where does it come from? was it rated to do 200% overclocking of GPU? How many devices are shared?

Let' hear it from Rasz, before we try few things.

[Sikator]

The powers supply used was a Corsair AX760 I would not think this was a problem. I have also tried the 780ti on 3 different motherboards. The psu has been tested with the 2 other PCI outputs (modular PSU) and tested on other GPU's no problem at all.

I have however tried to turn on the pc, then connect various psu cables to the GPU (780TI) Where it seemt like the x2 from the 6x2 made it do its 100% spin and not boot of death behaviour.

EDIT: The current gpu being used does NOT use the x2 from the 6x2 pins though.... It uses 2x6 pins. This couldn't possibly mean anything could it?

I could test again to see exactly which connected cables makes it do its death dance.

[Armadillo]

Please try the various combination BUT BUT on mundane clock only < 100%, no overclocking. 

[Sikator]

I will try it a in a few hours then! It has stock bios and no way of overclocking as far as i am aware so it should be fine

[Sikator]

Please try the various combination BUT BUT on mundane clock only < 100%, no overclocking. 

The results are in. Note that for some reason some configurations rarely doesn't make the gpu freak out every time, but after 1-2 tries it freaks out and then it sticks like that. The ones labeled Calm Never freaked out.

imgur was down so lets try imageshack: https://imageshack.com/a/u2Ul/1


EDIT: Also i was not able to get it recognized at all. even trugh nvflash listing. I do think last time i got it recognized because i used a separate GPU on the side and not the IGPU (cPU Graphics)

EDIT#2: Added a missing pic to the album on imageshack.

[Armadillo]

I am thinking;

The 780TI needs minimum 600W at current rating at least 42 amps at 12V. Considering the corsair AX760 rated 760W continuous, so I think the PS, that's about it, not to power anything else power hungry stuff...  I think only plug into the 2 X [4 + 4] CPU slots of the corsair taking in only 12 V.

But regardless, why it is not recognise at all?  emm.......

[1] not recognise
[2] freeze up the pc.

Some experts here may provide better advise; 

[Sikator]

I am thinking;

The 780TI needs minimum 600W at current rating at least 42 amps at 12V. Considering the corsair AX760 rated 760W continuous, so I think the PS, that's about it, not to power anything else power hungry stuff...  I think only plug into the 2 X [4 + 4] CPU slots of the corsair taking in only 12 V.

But regardless, why it is not recognise at all?  emm.......

[1] not recognise
[2] freeze up the pc.

Some experts here may provide better advise; 

Tbh those numbers from vendors are with the whole system and usually (always) a bit oversaturated Go to 1min in vid 780ti +some on a 450W

 

[Sikator]

It is really unfortunate, I really thought it would be a dead mosfet. How would it act if a cap was dead you think? These caps don't seem like the ones that blow when they die?

[Sikator]

Lets hope, Maybe Rasz. I really appreciate your time spent on this Armadillo. 

[Rasz]

Interesting, first, you ment like this right?

and i would test like this on all four top SIC780's?
http://image.prntscr.com/image/8f98b1ca72fc410ea12957c6f3deee41.png

yes, but remember this is view from the bottom = mirrored

EDIT: If so, Most gives 3.3 or 3.0 measured with 20k on my multimeter. Some configurations gave 0.00 but so did all the others in the same configurations.

there is only one configuration, pins are either PGND or VIN, see attachment
3 ohm between Vin and ground is BAD, Kohms sounds better - this would keep raising as the caps are being charged

There should be around ~1 ohm between SW and ground, this is GPU resistance thru the coil. I dont know the exact resistance of 780TI chip, and am too lazy to power down my own computer to check , but the GTX260 that just happens to sit next to me is 0.8 ohm, while GTS450 is 2 ohm, so expect something low, but not zero. GTX 1080 is <0.1 ohm so requires very good multimeter (perhaps even bench meter and 4 wire setup).

Now the 4901NF confuses me a bit aswell. the two 4901NF's do act different in mesurements as showed futher up.

thats because top one is supplied from PCIE, and the bottom one from external connector, different cap configurations etc. Your measurements make no sense without units, there is no way they are all in ohms +  Armadillo made you measure supply rails, not the output ;-). Output of 4901NF is on the bottom, you have to measure directly on the coil, or on that small black capacitor on its left (while looking at the card like in https://3dnews.ru/assets/external/illustrations/2014/03/27/814253/11_pal78ti_pcb_big.jpg )
http://www.onsemi.com/pub/Collateral/NTMFD4901NF-D.PDF


http://imageshack.com/i/plLbI9fKj
http://imageshack.com/i/poZSBIA8j
are same picture different subtitle ;-)


Lets start from the top. measure 3 fuses, the yellow/brown elements marked R or X, L16 L17 L18 I think, and R51 R52 R53 power monitoring resistors. While you are at it measure all the coils (grey cubes).
Then stick black probe on the metal bracket of the card or blue GND on my attachment and measure SIC780 Vin and SW with red probe in resistance mode. Pay attention to the units (ohms or Kohms), and if the reading is immediately stable or rising.



btw you dont have to plug those small x2 power connectors, at least not in the connector closer to the gpu, photos on the net show most Palit 780s dont even have full 8x4 socket there.
https://www.ekwb.com/configurator/upload/pictures/Palit-GeForce-GTX-780-Ti-JetStream-3GB-GDDR5-(NE5X78TH10FBJ)-PCB_54863.jpg
edit: older cards I have on hand all detected external plug by checking middle ground pin
Just for reference GTX 260 will print same message with no external power and will start with full tilt fans and black screen/no boot when it fails to detect both power connectors. I used to run it with only one plugged + paperclip shorting middle ground to its neighbor in the empty external power socket because my power supply had only one 6x2 plug. I do not recommend that very prone to frying whole computer by accident
Your card doesnt seem to use same detection method and instead measures power drawn from external sockets.
Either one of the mosfets in the middle is shorting supply,or one rail went completely open, or its a fried driver, or fuse, or power resistor, lots of stuff might of went bad. But getting the message with no external power means GPU and RAM are fine and card is fixable.

[Armadillo]

I suspect one of the mosfets is shorted between Vin and ground, and luckily its the one powered from external socket = you can still boot without external power and get usable picture with "please plug blabla" message.

We did suspected a direct short between Vin and ground, but I think it's a positive move, a way forward to suspect the switch as you have described. 

[Sikator]

Lets start from the top. measure 3 fuses, the yellow/brown elements marked R or X, L16 L17 L18 I think, and R51 R52 R53 power monitoring resistors. While you are at it measure all the coils (grey cubes).
Then stick black probe on the metal bracket of the card or blue GND on my attachment and measure SIC780 Vin and SW with red probe in resistance mode. Pay attention to the units (ohms or Kohms), and if the reading is immediately stable or rising.

While my multimeter was set to 200M (i don't know exactly how these numbers make eachother. in that case 1 would be 1mohm i guess?)

L16 and L17 measured 1.0 Mohms? in any orientation and L18 measured 1.1 Mohms? on both but with red on the right it kept going. highest i saw was 88.0 Mohms? still with multimeter set to 200M. like in this pic:

the chokes also measured 1.0
All of these above were measured alone (just one component)

If you wanted me to measure in any of the other ohm options on my multimeter then the only one that gave result was "200" (bottom on of the ohm options on my multimeter) And then everything measured 00.2-00.3Ohms


Note: Just found out that everything measures differently with black on a ground point on the PCB itself. so i will include some of those aswell in pics.
i am confused but ill keep going. i feel like i am blind folded. But Thank you for still sticking with me.

thats because top one is supplied from PCIE, and the bottom one from external connector, different cap configurations etc. Your measurements make no sense without units, there is no way they are all in ohms +  Armadillo made you measure supply rails, not the output ;-). Output of 4901NF is on the bottom, you have to measure directly on the coil, or on that small black capacitor on its left (while looking at the card like in https://3dnews.ru/assets/external/illustrations/2014/03/27/814253/11_pal78ti_pcb_big.jpg )
http://www.onsemi.com/pub/Collateral/NTMFD4901NF-D.PDF

I took "small black capacitor" as in the tiny things right next to them. Hope im right. i also did the other components at the same time for info in case you needed it in this pic:



Now the Coils which i assume are these pic:


Here is my multimeter. And what i mean with "measuring with 200k" for example is when ive choosen that on my multimeter.



I will try to do the SIC780's now but ill post this so i don't mess it up or make things more confusing.

[Rasz]

This makes no sense. There is no way for L18 to measure in megaohms, that is the only fuse almost guaranteed to work, otherwise you wouldnt get the "please connect" message. Im baffled
Either you made a mistake or found blown fuse, recheck it.

>I took "small black capacitor" as in the tiny things

this could be a winner ...if you actually measured what I asked one is not like the others on that pic, top SIC780 doesnt match the rest. BUT I didnt ask you to measure on small caps left of the SIC780  + you blue boxed a bunch of components and pins together so there is no telling what you were measuring

we need readings between green lines on attachment, 3 measurements per mosfet, 1 per L/R components.
your "200K" actually shows a picture with 200 selected  this (and earlier 2 identical photos with different descriptions) is a point when I start suspecting trolling. Im a terrible person to be teaching basics, I get frustrated quickly  . You need a multimeter / measuring tutorial, im incapable of teaching that 

[Sikator]

we need readings between green lines on attachment, 3 measurements per mosfet, 1 per L/R components.
your "200K" actually shows a picture with 200 selected  this (and earlier 2 identical photos with different descriptions) is a point when I start suspecting trolling. Im a terrible person to be teaching basics, I get frustrated quickly  . You need a multimeter / measuring tutorial, im incapable of teaching that 

I am sorry.

The pic was mainly for you to see what options i had so you knew what i ment with 200k, 200 or even 2000 further up in the thread. when you said "Your measurements make no sense without units" and the two different pictures i explained. i labeled them on my phone right after i took the pics. After uploading i looked trough them fast and didnt see the one i then edited with paint then uploaded.



You can be sure im not trolling, i would not sit 1.5 hours measuring and trying to learn for that, I am really trying my best to give you the info i can. and i will try again. I feel like a blind person being led trough this stuff.

i am going to do them again. exactly as you asked now. Some of these resistors and fuses i've done and while writing this i still cant make them read more than 0.02 ohms or 1.0Mohms. and all other options under OHM on my multimeter gives 000 no matter orientaton. this is why i am so confused. i got more numbers from using ground as i did on my last post, so i thought that had to be right.

This goes for R51, R52, R53 and L17. so essentially open/short?
L18 is the only one with any sign of life. Thats the bottom one but it is also confusing, here are two readings i can get from it:

and

I am doing the SIC780's now but don't know if i have time to finnish it so i'll post what i've got so far. Regarding the 4901NF's i don't understand which is ground trough the datasheet. That is why i measured both ways.





I really do appreciate the help here Rasz.

[Chalcogenide]

The red components with an X on top are most likely fuses, so the one reading high resistance is likely dead. You should verify everything connected on the line that the fuse was protecting; if nothing seems wrong, replace it with a similarly rated fuse (I couldn't find a datasheet, maybe someone more expert will be able to help) and hope for the best.

[Rasz]

i am going to do them again. exactly as you asked now. Some of these resistors and fuses i've done and while writing this i still cant make them read more than 0.02 ohms or 1.0Mohms. and all other options under OHM on my multimeter gives 000 no matter orientaton. this is why i am so confused. i got more numbers from using ground as i did on my last post, so i thought that had to be right.

you dont have to make them read anything specific, the point of measuring is getting a readout. 0 is absolutely fine, fuse should read 0, resistors should read very close to 0 (0.005 ohm).

L18 is the only one with any sign of life.

thats GREAT!

The red components with an X on top are most likely fuses, so the one reading high resistance is likely dead.

yes. This fuse is blown/open circuit.
I think I know whats going on now, pictures misled me. I was assuming bottom fuse PCIE and top ones external.
Measure red-green and red-blue Edit:on first picture. Im guessing red-green will be 0. I was assuming from the start to be the other way around. Now it all makes sense, even the weird top SIC780 reading.
btw its easier to measure using Diode mode - 8th setting on your multimeter, the one with speaker icon, it makes beeping noise when shorted/detecting connection.

most sane Google fuse data I found said
R = 3.5A
X = 8A
~240W total makes sense for 780TI, Nvidia claims 250W TDP. This also explains why you blew up your card, even GTX770 can draw 230W, and 780TI is 2x the size. Unlocking power limit made the card stop throtting, this option is for serious "enthusiasts", people who solder things like EVGA EPower to their cards.
http://extreme.pcgameshardware.de/attachments/521480d1329751996-vmod-extreme-evga-epower-board-untouchables-im-test-complete_2.jpg

Best case scenario you only blew a fuse, $1 and you will be back in action.

Regarding the 4901NF's

argh, i knew I should of used different colors , there are THREE lines there, now in more colors

I am sorry.

no, I am sorry! Im terrible with a/children b/explaining very basic stuff. My personal worst was "tutoring" networking to 4th semester CS girl. I started by showing something cool - l33t networking hacking (aka script kiddie tools like kain&abel, this was ~2004?), but It soon dawned on me 2 years of SC apparently wasnt enough for her to learn absolute fundamentals like osi model, or whats a mac address An hour with me made her drop out of college .

[Sikator]

you dont have to make them read anything specific, the point of measuring is getting a readout. 0 is absolutely fine, fuse should read 0, resistors should read very close to 0 (0.005 ohm).

Now i understand it a bit more
I am learning a fair bit here it feels like

Measure red-green and red-blue Edit:on first picture.

First pic
Red + Green = 002 (diode mode)
Red + Blue = Does get numbers but in a split second goes above what my multimeter can read. "1   . " (diode mode)

Second pic
Red = 146 with Red on the tiny thing and 103 with Red on the 4901NF Chip. (diode mode)
Green = 522 with Red on top. And above what it can read with Red on bottom. (diode mode)
Blue = 550 with Red on the tiny thing.  And above what it can read with Red on the 4901NF Chip. (diode mode)

most sane Google fuse data I found said
R = 3.5A
X = 8A
~240W total makes sense for 780TI, Nvidia claims 250W TDP. This also explains why you blew up your card, even GTX770 can draw 230W, and 780TI is 2x the size. Unlocking power limit made the card stop throtting, this option is for serious "enthusiasts", people who solder things like EVGA EPower to their cards.
http://extreme.pcgameshardware.de/attachments/521480d1329751996-vmod-extreme-evga-epower-board-untouchables-im-test-complete_2.jpg

Best case scenario you only blew a fuse, $1 and you will be back in action.

  Trying not to get too hyped   

And yeah, i knew but being stupid is more fun some times until you fail i guess. This gpu actually got really good overclocks, even at decent temps. so i felt there was more to get, And there were until i underestimated and blew it at 200%.



After googeling a bit, i read that Palit gives exact component specs etc if asked. And i got a positive answer when i sent them an email.

"Dear customer
 
Thank you for the mail.
 
If you need some components’ SPEC please kindly provide below information.
1. S/N of the graphic card, you can find the S/N from the product box or the sticker in back of the graphic card.
2. Big & clear pictures of the PCB part of the graphic card and please use color circle to point out what components you need the SPEC.
We will provide the components’ SPEC accordingly.
 
Please kindly be notified.
Thanks."
Will be good i guess for atleast when/if i need to order one of those unlabeled parts

no, I am sorry! Im terrible with a/children b/explaining very basic stuff. My personal worst was "tutoring" networking to 4th semester CS girl. I started by showing something cool - l33t networking hacking (aka script kiddie tools like kain&abel, this was ~2004?), but It soon dawned on me 2 years of SC apparently wasnt enough for her to learn absolute fundamentals like osi model, or whats a mac address An hour with me made her drop out of college .

Imo, trying to learn is useless to begin with if there isn't enough interest in the subject. (if i got the scenario right)