GTX 780tI Repair attempt...

[dicky96]

Re: GTX 780tI Repair attempt...

Updates.....
PALIT GTX 780ti

I removed all the Video RAM. Due to practicality, and the way my workshop is layed out, I actually used the Kada 853B to preheat the board to about 120C and the Quick 861DW to remove the chips. All went well apart from I slipped with one RAM chip on the vacuum pickup and took out 4 nearby SMD resistors/capacitors accidentally

But that is not a disaster - I can easily work out the values and replace them, the same components repeat for each pair of RAM chips.

I cleaned up the PCB and there are no damaged pads, so that's good.. I tested for the short on Vram after removing 4, then 8 chips and the short was still there. With all chips removed it is gone and I have 110 ohms - same as my 'kinda' working board

As the VRAM FETs were faulty and the inductors are currently removed, I then powered up VRAM from my bench supply at 1.5V - something draws about 2.7A and I am pretty sure that is the GPU as it gets a bit warm. I then powered up the motherboard from ATX supply and I have a sort of result!

Basically the PC now beeps to show it detected the graphics card, and my monitor comes on so it must have video sync - but I have a blank screen. That would seem reasonable as I have no video ram on the graphics card now...... but...

I saw this video

https://www.youtube.com/watch?v=vDyTtcKndug&t=601s

I am basically doing the same thing as this guy, with similar reuslts on the current drawn from the bench PSU (I think his is a GTX780 and mine is the 780ti which makes sense mine would draw a bit more power) but he gets a picture with no video ram and I don't. Watch from 5 mins onward to see this. Unfortunately the video is in Russian and I don't understand that at all, but in the end he fixes his card, though i am not sure if he also reballs or replaces the GPU.

Does anyone here understand Russian and can explain what this guy is saying? And how he gets video output with no video ram?

I could order some Video Ram and FETs for this card, plus replace the few missing SMDs caused by my slightly dodgy desoldering.. It would cost about 20 euros in parts to continue with this one. Should I? I am tempted too, just out of curiosity.


PHANTOM GTX780
Kinda working now. I need to replace those missing 6 SMDs see if it gets rid of the artifacts in the video. If that don't fix it, is there any way to determine what is causing these? Heating / Cooling the GPU, RAM to see if it affects the picture?

GIGABYTE GTX780ti
Waiting for Stencil to replace the GPU. Needs two SMD Capacitors and a Tantalum 470uF 3V. Parts ordered.

Gigabyte GTX770
Will look at this one next see why a FET blew an hole in itself when I see no shorts

[dicky96]

Re: GTX 780tI Repair attempt...

And now onto the Gigabyte GTX 770, GV-770OC-4GD

Someone has been messing with this one, there are three electrolytics missing plus a tantalum from the other side. Also a chip is missing (I can see where they cut the legs off to remove it). I don't know what the chip is, but next to it is a LM358L and these look like they may have been a pair looking at the PCB layout. Alternative scenario is something hit this part of the PCB very hard, as it looks possible the chip was smashed off the PCB and the missing electrolytics which are right next to it had the legs still attached to the PCB just no body.

It is possible other parts are missing. In fairness I only paid 2 euros for it a the car boot sale.

I think the only chance with this one would be to buy another card of the same model (working or spares/repair) then compare the two in the hope to get two working cards ,as I don't know for sure what may be missing..

I took some resistance measurements, Vcore is 6R3 and Vram is 44R - which looks kinda OK compared to the other cards, considering this one has 16 RAM chips, 8 each side and the GTX780 cards had 12 RAM chips and read 110R

The blown FET is short Gate- Drain I don't know why that has gone bang. Could be related to those broken parts?

There is something *very odd* about the missing capacitors as I can't figure out where they connect to. I thought they should be on Vram because they are right next to the inductors but I found two capacitors for that supply rail further across the board. They don't connect to 12V in on the 8 way connectors or the PCIe 12V either. Also they are not in parallel with the three electrolytics just below them. I can't make sense of it unless I am missing something obvious. see pics.

I think the Vcore/Vram resistances look OK. What do you guys think?

Next I will have a look at the Phantom again, try replacing the six missing SMDs from the Palit see if those video artifacts go away.



I'm still hoping for some advise on the Palit, is it worth attempting to replace the RAM chips? These were obviously short circuit.


Rich

[momaka]

Re: GTX 780tI Repair attempt...

Updates.....
PALIT GTX 780ti

I removed all the Video RAM.
...
I tested for the short on Vram after removing 4, then 8 chips and the short was still there. With all chips removed it is gone and I have 110 ohms - same as my 'kinda' working board

Wow, I'm actually really surprised by that result.
As mentioned, I have never seen RAM chips go short-circuit - especially without overheating when powered up. I really thought it would have been the GPU shorted on the RAM supply rail, but I guess that goes to show that RAM chips can also go bad. That said, I wonder if the MOSFET on the RAM supply shorted and killed the RAM chips, or the other way around. I have a HIS Radeon HD3870, whose PCB conveniently has jumper links separating the RAM supply into two traces: one for the RAM chips and one for the GPU. Removing these, I was able to identify that the short is stemming from the GPU tracks and not the RAM. And as mentioned, I've seen this on other ATI/AMD GPU chips as well (including all the way back to Xbox 360 GPUs.) So the results you got with your nVidia GPU definitely surprise me. But I guess nVidia and ATI/AMD can have different failure modes in that regard.

That said, I suggest you clean the removed RAM chips from your card and test each one to see which one is shorted... or at least to confirm that's where the short-circuit came from. Just measure between each RAM chip's Vdd and Vss pads.

As the VRAM FETs were faulty and the inductors are currently removed, I then powered up VRAM from my bench supply at 1.5V - something draws about 2.7A and I am pretty sure that is the GPU as it gets a bit warm. I then powered up the motherboard from ATX supply and I have a sort of result!

Basically the PC now beeps to show it detected the graphics card, and my monitor comes on so it must have video sync - but I have a blank screen. That would seem reasonable as I have no video ram on the graphics card now...... but...

Yup, I can confirm the GPU should detect and monitor power up without the RAM. I tried this on an old EVGA GeForce 7600 GT video card that had shorted RAM chips due to faulty caps. I removed the RAM chips and tried powering on the card without them. The only thing is, I don't remember if I got a black screen or a screen full of colorful random/noisy artifacts (I know I got both at some point, but I don't remember if one was with the GPU RAM supply disabled or whatever else I tried.) This was about 6 years ago, so I just don't have good memory what I tried on that GPU anymore.

I saw this video

https://www.youtube.com/watch?v=vDyTtcKndug&t=601s

I am basically doing the same thing as this guy, with similar reuslts on the current drawn from the bench PSU (I think his is a GTX780 and mine is the 780ti which makes sense mine would draw a bit more power) but he gets a picture with no video ram and I don't. Watch from 5 mins onward to see this. Unfortunately the video is in Russian and I don't understand that at all, but in the end he fixes his card, though i am not sure if he also reballs or replaces the GPU.

Does anyone here understand Russian and can explain what this guy is saying? And how he gets video output with no video ram?

I'll try to give it a watch later tonight (posting this from my workbench Pentium III laptop right now - no YouTube! ). Don't understand much Russian, but may be able to catch some words.

I could order some Video Ram and FETs for this card, plus replace the few missing SMDs caused by my slightly dodgy desoldering.. It would cost about 20 euros in parts to continue with this one. Should I? I am tempted too, just out of curiosity.

If you do determine that one or more of the RAM chips really are shorted (by cleaning and measuring the old RAM chips), then it may be worth it indeed - especially to satisfy the curiosity part.

PHANTOM GTX780
Kinda working now. I need to replace those missing 6 SMDs see if it gets rid of the artifacts in the video. If that don't fix it, is there any way to determine what is causing these? Heating / Cooling the GPU, RAM to see if it affects the picture?

Nice, good to see it's almost getting there working.
The artifacts look a bit random, suggesting perhaps noise on the RAM supply. So replacing the missing SMD parts may actually fix it.

If not, inspect all of the SMDs on the card for hit/crack damage - particularly, look for shorted small capacitors or open-circuit small resistors. I revived a Gigabyte Radeon HD6850 that had a shorted small ceramic cap. Not sure if from hit damage or by itself, but the card did have a few other missing SMD ceramic caps, so external damage is not out of question. I made a thread about it here:
https://www.badcaps.net/forum/showthread.php?t=81340

GIGABYTE GTX780ti
Waiting for Stencil to replace the GPU. Needs two SMD Capacitors and a Tantalum 470uF 3V. Parts ordered.

Fingers crossed it works.

[momaka]

Re: GTX 780tI Repair attempt...

And now onto the Gigabyte GTX 770, GV-770OC-4GD
...
Someone has been messing with this one, there are three electrolytics missing plus a tantalum from the other side. Also a chip is missing (I can see where they cut the legs off to remove it). I don't know what the chip is, but next to it is a LM358L and these look like they may have been a pair looking at the PCB layout. Alternative scenario is something hit this part of the PCB very hard, as it looks possible the chip was smashed off the PCB and the missing electrolytics which are right next to it had the legs still attached to the PCB just no body.

It is possible other parts are missing. In fairness I only paid 2 euros for it a the car boot sale.

I think the only chance with this one would be to buy another card of the same model (working or spares/repair) then compare the two in the hope to get two working cards ,as I don't know for sure what may be missing..

You may be able to find high-res pictures online of that PCB and determine what the missing chip is.

I suspect the damage to those missing parts was from hit damage. And there is also a chance the GPU or something else could have gone bad before that, thus making the original owner not care and not store it properly (or discarded it and that's when the damage happened.) So that's probably going to be a multi-part repair, even after getting all of the right parts installed again.

I took some resistance measurements, Vcore is 6R3 and Vram is 44R - which looks kinda OK compared to the other cards, considering this one has 16 RAM chips, 8 each side and the GTX780 cards had 12 RAM chips and read 110R
...
I think the Vcore/Vram resistances look OK. What do you guys think?

Looks alright to me as well. At least nothing obviously bad.

The blown FET is short Gate- Drain I don't know why that has gone bang. Could be related to those broken parts?

Or the other way around.
FET blew and original owner threw the card away, damaging it in the process??
Sometimes, we may never know. :\

Next I will have a look at the Phantom again, try replacing the six missing SMDs from the Palit see if those video artifacts go away.

Sounds like a plan.

I'm still hoping for some advise on the Palit, is it worth attempting to replace the RAM chips?

Well, all the time and parts you put into that card... probably not worth it from a business standpoint. But for curiosity, learning purpose... and fun?? it probably is.

Also, I think when you repair something like this that no one else thought is possible... I feel that it gives you more confidence at the end of the day too.

[dicky96]

Re: GTX 780tI Repair attempt...

PHANTOM
OK so i replaced the six missing SMD capacitors by taking them from the Palit. I also made a note of the values.

From the resistance across them, one is on Vcore, one is on Vram and the other four are on the 3.3V rail from PCIe

Unfortunately this didn't make any difference, I still get those dashed lines on the video output.

So I am not sure how to diagnose that further. As an aside - how do I power up the graphics card for more than a few seconds with no heat sink mounted on the GPU, so I can actually probe around it?


PALIT
I couldn't find a datasheet for the Video RAM, H5GQ2H24AFR but it was easy to find out a pin that is Vram and a pin that is Ground from checking on the PCB. All of the RAM are short - though they vary a bit, the lowest one I found was 0.6 ohm and the highest 1.7 ohms. Most were around 1 ohm. So I will order 12 new RAM chips today for this board and see if it fixes it.

GTX770
Actually I can tell you which way round the FET blew. When I first got the card a few months ago I plugged it in to see what it did. Yes I should have spotted the missing parts first! Anyway the first couple of times I turned on the ATX PSU nothing happened, and the third or fourth time that FET sparked and a bit blew out of it. So the FET blew last. Interesting it is short GATE-DRAIN but not SOURCE-DRAIN which is what I usually see.

I'm looking for another GTX770 card of the same type Faulty or not - for comparison.

So right now I can't really do anymore with any of these until I get some parts which will probably take a month at least from Aliexpress.



Rich

[momaka]

Re: GTX 780tI Repair attempt...

PHANTOM
OK so i replaced the six missing SMD capacitors by taking them from the Palit. I also made a note of the values.

From the resistance across them, one is on Vcore, one is on Vram and the other four are on the 3.3V rail from PCIe

Unfortunately this didn't make any difference, I still get those dashed lines on the video output.

So I am not sure how to diagnose that further.

Check resistance across all small resistors near the RAM chips and compare them to the other card(s). Do the same with the small ceramic caps near/under the RAM chips. Not all of them are connected to Vram, Vcore, or Vtt. Some are used for signal coupling, like the one I had shorted on my Radeon HD6850. There are very few of these caps, but it only takes one to give bad image.
... yeah... that's going to be a hella lot of part testing, I know. But it's the only way to be sure.

On a related note, I'm currently playing with a HIS Radeon HD4870 that had a couple of small MLCC caps knocked next to the PCI-E connector. According to the seller, that is very likely all it needed to get it going.
Welp, I fixed those and card still acted dead-ish (all voltage rails present, GPU HS heated up and consumed normal power for a HD4870... but not video.) Then I took a few measurements on the rest of the original MLCC caps by the PCI-E and found one that reads very low capacitance - around 50 nF instead of ~100 nF, which is what those caps are supposed to be. Measure directly on the cap and got close to 100 nF. Measured on the traces again, and got low capacitance. Then measured on the cap and got infinite capacitance (short-circuit / leakage?) Then measured on the cap again and got different capacitance again. So it seemed putting pressure on the cap made it go bad.
Replaced that cap, along with 2 others that did the same thing. Put the GPU back in and this time the Num Lock on the KB came up and stayed lit (previously, it would light up and then go out, but not freeze, indicating GPU possibly detecting and then error-ing out.) But it looks like I have more of those caps by the PCI-E connector bad, because as I moved the card up and down on the motherboard to flex it near the PCI-E slot, the POST behavior changed between old way with the faulty caps I was initially getting and the partial boot-up with the replaced caps. And I could change exactly the behavior whether I pushed up or down on the card. So there are probably more of those PCI-E caps that are faulty.

Thus, moral of the story is, you really have to check everything.

As an aside - how do I power up the graphics card for more than a few seconds with no heat sink mounted on the GPU, so I can actually probe around it?

Wait, you've been powering your card(s) this whole time without a heatsink installed?

That's a recipe for killing any silicon die!

The only way you can power the GPU and still have an easy way to probe around it is to use a small water block and liquid cooling. Or maybe... and this may not work or be easy to adapt... an old PS3 phat heatsink attached to the GPU. Those use copper heatpipes and move the heat to fins on the side, so that may give you enough clearance around the GPU chip.

PALIT
I couldn't find a datasheet for the Video RAM, H5GQ2H24AFR but it was easy to find out a pin that is Vram and a pin that is Ground from checking on the PCB. All of the RAM are short - though they vary a bit, the lowest one I found was 0.6 ohm and the highest 1.7 ohms. Most were around 1 ohm. So I will order 12 new RAM chips today for this board and see if it fixes it.

Interesting.
I thought only one or two of those chips would have shorted, saving the rest from going bad. Makes me wonder why they all shorted in the first place.

GTX770
Actually I can tell you which way round the FET blew. When I first got the card a few months ago I plugged it in to see what it did. Yes I should have spotted the missing parts first! Anyway the first couple of times I turned on the ATX PSU nothing happened, and the third or fourth time that FET sparked and a bit blew out of it. So the FET blew last. Interesting it is short GATE-DRAIN but not SOURCE-DRAIN which is what I usually see.

I've seen MOSFETs go short-circuit on Gate-Drain and Gate-Source. Usually a lot more rare, but not impossible. The bad thing about Gate shorting to anything is the controller/driver for that MOSFET may (likely) need to be replaced too.

Also, when you say "nothing happened" when you turned On the ATX PSU... do you mean the system tried powering on for a second and then turned off? Or powered On, but just no video. The former would indicate PSU detected short-circuit and shut-down... which means either the MOSFET was still bad before this or the RAM was shorted and MOSFET not bad yet.

So right now I can't really do anymore with any of these until I get some parts which will probably take a month at least from Aliexpress.

Well, that may not be a bad thing.
Sometimes, I find it good to pull my head away from a project and work on something else. In fact, this has allowed me a good number of times to analyze a problem differently and find what was actually wrong. It's as if thoughts have to "simmer" in my head on their own to get to the right solution, sometimes.

[dicky96]

Re: GTX 780tI Repair attempt...

I was powering on the card only for a few seconds with my fingers against the GPU to check the temperature. As soon as I saw an image I turned off again. It only got warm, not hot. I did myself think a water cooling block is something i need to get, I was just hoping for a cheaper way. Previously i had no Vcore or Vram or PLL voltage rail (what does that do by the way? PLL = phase locked loop = oscillator?) on this card so the GPU was cold. The others don't giver an image but again I only power on for a few seconds to take a voltage reading or see if the GPU detects, and use the same technique of fingers on GPU. Never has one got hot enough to be uncomfortable to touch.


I put it back together to see what it does - those dashed lines are like about an inch long and almost always dark blue, sometimes cyan or green - against white they are yellow

They are only on the top half of the windows desktop, and not always in the same place.. Images are fine it is just the windows 'background' area you see the dashed lines. When an image appears it kinda erases and dahed lines that were there. I suspect this is RAM problem as it is only in certain areas of the screen but have no experience to be sure. I will try get some video of what I mean after the weekend

So Windows runs basically OK, however the PC always crashes (black screen) once I install the GTX 780 drivers - I tried a few times with Windows 10 64 bit, then put another drive on with XP 32 bit but the same thing happens.

I will have a look like you said at MLCCs

Cheers

[dicky96]

Re: GTX 780tI Repair attempt...

Ahaaa

I read a useful tip on another forum post here

To understand a russian youtube video

https://www.youtube.com/watch?v=vDyTtcKndug

Click on CC icon on youtube player to turn on closed caption, then go to settings and choose translate - this worked well enough for me to understand this guy. And there was a LOT of useful info in the video. I now have some things to try before i order RAM chips for the PALIT card :-)

[dicky96]

Re: GTX 780tI Repair attempt...

OK So I had a look at the resistance on all the DQ lines on the PALIT which has the RAM removed.

I can't find a ball-out for H5GQ2H24AFR or even for GDDR5 though I think probably it is a standard layout. If anyone has that datasheet i would much appreciate it.

However in the Russian video at 2min 42 he shows which balls to check, and it is good enough to work from. There are 64 of them - though I read that GDD5 is 32 bit so I am not sure why that is??

Anyway, He says that if memory controller in the GPU is good you should read the same resistance on all these pins. On mine I didn't bother checking all channels (I think this card has 6 channels with two chips each) but I didn't need to because on M1 M5 and M8 chips - which I chose at random - I see I have around 560Kb on most of the DQ pins and 80Kb or less on some others. A few reads low at 9.8K

So from that, and the fact my GPU detects but gives no image without Video RAM whereas apparently I should see an image if the GPU is working, leads me to the conclusion my GPU is damaged and there would be no point replacing the Video RAM

The GPU are quite expensive to buy - around 75 euros on Ali but I did find a supplier at 45 euros which would mean I could at least get my money back if I fix this card, and probably even make a small profit at current used pricing - and it would be a useful learning experience, which is something you can't really put a price on.

I think I will put it on one side a little while. If I manage to reball and fit the Gigabyte GPU and that card is then working, I may decide to try fix this one also

I'm vacillating over removing all the RAM from the Phantom (the card with those lines on the picture) and checking the DQ resistances on that one also - to see if I can prove whether I have artifacts from a faulty GPU or from faulty GDDR


Rich

[piernov]

Re: GTX 780tI Repair attempt...

GDDR5 indeed has a 32-bit data bus, so 32 DQ pins, but you also have the address bus, the control bus and even a separate bus for CRC, and the clocks.

Random datasheet of a GDDR5 chip attached (from )
There might be different packages depending on the density of the chip, haven't checked.

I find it somewhat surprising to measure in resistance mode. Personally I'd use diode mode as usual on signal lines. But I'd expect all the signals in given set (eg. the 32 DQ pins of the data bus) to show a very close value on the multimeter since they should have the same driver inside the memory controller. The value may be different from one set to another (since for example a differential clock line will have a different driver than a single-ended data line).

Also a GPU could fail without showing any weird measurement on the RAM bus using the multimeter.

[SMDFlea]

Re: GTX 780tI Repair attempt...

H5GQ2H24AFR Datasheet .The ballout starts on page 156 .Top View (as seen thru package)

.

[dicky96]

Re: GTX 780tI Repair attempt...

Ahh thanks SMDFlea - that's excellent!

OK so I can now see what signals I am probing, here is an example of what I find on this Graphics card. I picked 2 random chips and checked all signals marked in green on the ballout diagram

GDDR IC M5
Byte 0
DQ3= 55K
DQ7 = 122K
All other balls Byte 0,1,2,3 read 560K-600K


GDDR IC M7
Byte 1
DQ14 = 88K

Byte 2
DB12# = 51K
All others Byte 0,1,2,3 560K-600K


It is obvious from the PCB layout that M5/M6 belong to one channel and M7/M8 to another - I am not sure of the way the layout is sequenced but if we go round in a clockwise direction lets say we have channels 0-5 in sequence and these are channels 2 and 3

So it is clear that this GPU has problems on Byte 0 of channel 2, but byte 0 checks OK on channel 3 where we have problems with Byte 2 and 3 instead

It looks like I stumbled on a good diagnostic method here to prove if the GPU is bad before going to the expense of changing the GDDR Ram

Obviously this is not new knowledge, but was I the only on here who didn't know it?

What I will do now is remove the GPU from the PALIT and see if I can find any differences with the supposed good one I already removed from the Gigabyte.

This may assist if I decide to buy a replacement GPU from Aliexpress in as much as I can test it for some problems before fitting it. If i can spot differences between the two GPU that I already have, then I am certain I will have to order one because the curiosity will get to me too much

[dicky96]

Re: GTX 780tI Repair attempt...

I need to put these all on one side for a little while now until the GPU stencil arrives

I ended up removing the GPU from all three cards

The Gigabyte hopefully will be OK (short was MLCC not GPU)

The Phantom - I watched another video from the russian guy who has exactly the same sort of artifacts on a GTX780 that I am seeing. He fixed that by reflowing the GPU (using flux and enough heat to melt the balls) and it then worked.

I decided to do the same but accidentally pressed the wrong button on my semi auto BGA rework machine and it ended up removing the GPU instead of reflowing it LOL. I now have the PCB and BGA cleaned up ready to reball and refit

The Palit has a faulty GPU - determined by resistance measurements of memory controller buffers. If the other two are successful I will probably order a GPU and RAM for this one - parts cost about 65 euro but the card is worth about 120 euro, plus it is good experience which you can't really put a price on

The GTX770 4Gb I'm still waiting to find another of the same make/model to compare. There was one in auction on ebay yesterday but it went for more than I wanted to pay (and more than the average sold price which I would have paid). I'll bide my time and I am sure another will turn up

So for now this experiment is sleeping..... but I'll be back, as they say

[dicky96]

Re: GTX 780tI Repair attempt...

mrbrandin
Unfortuntely your technique would not have halped me with any of these GPUs. One had a shorted capacitor on the GPU Vcore supply (what could not be tracked down without removing e GPU first),

Another, none of the VRMs were running because of a broken trace or missing component on the 3.3V PCI voltage supply. Once I did fix that I had artifacts.

The Third had a faulty GPU, which was proven by removing all the VRAM and then measuring resistances on the internal VRAM buffers in the GPU by connecting to the balls under the RAM BGA's

I am all for putting the card in the oven as a final attempt - but what I don't understand (and makes me angry at times) is some well known 'famous, even' youtube channels, PC forums and the like, telling people to do this without making at least some basic checks first - like are the voltage rails present and do you have a short on any of them. Because if they aren't present, or you do have shorts, then how is the oven supposed to help?

And to do that you only need a €5 multimeter! FFS.