[SOLVED] Trying to repair a MSI 1080 Ti Armor

Feb 6, 2021
10
1
15
Hi there,

I'm new here trying to dive right into an electronics problem on my MSI 1080 ti armor. Maybe you can help me or point me to a different forum?

The card is 99% like this MSI GTX 1080 Ti Gaming X PCB:
View: https://www.youtube.com/watch?v=kj57W3vTqoQ


The card suddenly stopped working without any visible life sign. So I took it apart and replaced one defective "R005" resistor. Because maybe it just ran hot in that area.

Still the card won't boot.

So what I figured out is that there are many basic electronics parts that are needed for the card to boot properly. Like 1,8V for the bios and basic memory parts. But the 1,8V aren't present. They are provided by a AOZ1267QI-01 chip which is disabled because its "ENABLE" input is "LOW".
  • I am sure that all the 12V rails from the power supply are available where they should be
  • as well as 3,3V and 12V from the PCIE port. I can measure them everywhere where they may be needed.
  • Also the power supervision chip "INA3221" has all 12V rails supervised and outputs a "HIGH" on its "POWER GOOD" output. which is good.
  • all power DMOS / switches of the VRM seem to be OK
This "POWER GOOD" signal is connected to some "ENABLE" nets of some parts via a 470 Ohm resistor, so this should force them "HIGH" as well. But they aren't This means that
  • either there are more "POWER GOOD" signals which are forcing the "ENABLE" pins to "LOW"
  • or there is a short to ground somewhere on any of the parts connected together on the "ENABLE" net
Does anybody have a clue what to do next? I could desolder any suspicious part, but that would be a lot of stress for all the parts and PCB.


One more thing: Theres a funny part that I can't identify. It must be some sort of DCDC or power supervision. It is component U27 and has a "ADPJ" marking in a SOT23-8 housing. I've seen other 1080 TI boards where there was a similar component with "ADPG" marking. Any ideas here?

I'm really thankful for any hints/help.

Cheers
Axel
 
Solution
Update: Probably bad news (for my card).

A friend of mine needed to have his thermal paste renewed for his 1080 TI. Luckily he has the MSI 1080 Ti Gaming X from the video above.
So I helped him out and measured some resistance values before and after VRM modules.
I didn't know that memory power supply is completely seperated from the GPU supply. So I could have found that something's wrong earlier.
My Memory power supply is ~ 0,0x Ohms. Thats bad.


(Resistances measured to GND)Intact 1080 TI Gaming XMy Broken 1080 TI Armor
PEX / PLL Output80-100 Ohms100 Ohms
5V Rail (MP1475)Cycling value7 - 15 Ohms
Memory VRM...
Check the logic element (AND gate). Verify by removing it and inserting a pull-up resistor.
Would that help? The answer is in the last image, the first two are for having a bigger picture...


One more thing: Theres a funny part that I can't identify. It must be some sort of DCDC or power supervision. It is component U27 and has a "ADPJ" marking in a SOT23-8 housing. I've seen other 1080 TI boards where there was a similar component with "ADPG" marking. Any ideas here?
Step Down converter, MP1475, U501 (RT7296FGJ8F) in my first picture above.
 
Last edited:
  • Like
Reactions: Meffigold
Feb 6, 2021
10
1
15
Holy cow, thanks for the detailed help! :love:

I'll try your hints next week.

Is there a source for these schematics? Thats great stuff.

€dit:

Yes that MP1475 is probably the issue. This is what i measured at its pins last week. I added the pin description for the MP1475 and now it makes sense:

View: https://imgur.com/x1oMu3u


Its not working (forcing the ENABLE pins (its own PG pin) low) because its oviously not switching its 12V to 3,3V
 
Last edited:
Yes that MP1475 is probably the issue. This is what i measured at its pins last week. I added the pin description for the MP1475 and now it makes sense:

View: https://imgur.com/x1oMu3u


Its not working (forcing the ENABLE pins (its own PG pin) low) because its oviously not switching its 12V to 3,3V

  • The output should be 1.8v
  • The output is set by FB which is 0v in your case.
  • Lift it from the board and check inputs in open-circuit condition if you do suspect it.

Read the datasheet and check the card schematics I posted earlier, do not rush through conclusions.
I still suspect the AND gate... :)

hKzPMfE.png
 
Feb 6, 2021
10
1
15
I think our design slightly differs. I can't find a U505 anywhere. Maybe I'm just blind.

I also don't have that LDO 1,8v from the start of the video. I have the Footprint but the part is not placed.

€dit: OK I get it. Your schematics are from a different card with same architecture.

€dit2: 2,6 Ohms between SW and GND on the footprint - I think thats OK for a 1,8V rail. Either way about the AND gate -> shouldn't this thing turn on if all voltages are present? o_O

View: http://imgur.com/a/K5LVBHa
 
Last edited:
It should
Either way about the AND gate -> shouldn't this thing turn on if all voltages are present?
Yes. Page 2 of the datasheet has the logic table. I would just remove it, check VCC and the inputs, and (if all present) pull up the output by VCC with a resistor and check the EN input (leg 36) on the main power controller (u505 you asked about, uP9511P), controlled by NVDD_EN H/L. It might be easier if you find one of the R567/R571/C536 pads.
 
Feb 6, 2021
10
1
15
humm I really think that there are quite some differences between 1080 and 1080TI. I don't have a U505. But I'll try the MP1475 first and report back if it works :) Thanks very much so far!
 
Feb 6, 2021
10
1
15
The 8 phase controller is a NCP81274MNTXG.

I'm not sure if my power supply can handle stable 1,8V 1A+. So I'm waiting for spare parts for the APDJ (they cost like 30 cents).
 
Feb 6, 2021
10
1
15
i removed the resistor divider circuit for the ADPJ Feedback pin. it must be a much higher voltage than 1,8V. its Ohms 7,6k by 39,7k Ohms. probably ~5V for the GS voltage of the DMOS. Thats also what he says in the video.
 
Feb 6, 2021
10
1
15
I'm 99,5% sure that its 5V. I desoldered the parts, measured them. And its very close to the 5V Application example in the MP1475 datasheet (40,2k by 7,68k).

So I applied 5V with the card running but still no luck. That "power good" signal is still forced to 0V somewhere. Its really hard to figure out without 1080ti schematics. A lot of the the EN / Power good logic seems to be done by transistors on the backside. As I said, all different than in your schematic. I have a lot of transistors. not so much gate logic.

Another discovery:
050D150D (marking code) is a FDPC8016S. I first thought Q13 and Q14 were defective, but they intentionally shorted pin 1 to pins 2,5,6,7 with a 0 ohms resistor. They don't seem to need the high side switch. Just the low side one.

View: https://imgur.com/a/ilPvtxC
 
So it is logically disabled. You are moving forward. If the IC is doing good, you can safely put it back on. Look for the logic element that controls the switch transistors. It could be AND or OR. It would be the one that provides "low" signal functioning as the "power-good mux". Remove it from the board, measure inputs, measure output resistance, and (if sufficient) apply proper output voltage. Use IR camera if you have one along with a 100Ohm force-pull-up with low power and safe voltage values to find it.
 
Feb 6, 2021
10
1
15
I didn't write for some time because I'm waiting for a riser card for easier measuring and debugging.
I'm still on the logic enable debugging trail. Do you think its safe if I just manually pulled every logic / power input "high"?


Some other thoughts:

Every graphics card that I know has some sort of "no 12V rail connection" display output. So the user knows when he forgot to connect the big +12V power cables.
So 1,8V, 3,3V and 5V around the display connectors should be present for that. I only see the 3,3V.

Also my 5V rail indeed seems to have very low resistance... its only [EDIT!] ~7.3 Ohms. I think thats very low. But I'm not sure. Maybe there IS still an issue with the MOSFETS or driver ICs somewhere.

Damn I'm so happy to not have sold my cheap GTX 745 with 4 GB GDDR3. It really can do all of the work that the 1080ti did before. Even some fullHD low-mid settings gaming at 60fps.

Sorry its 7,3Ohms at 5V. I still think thats much?
 
Last edited:
I didn't write for some time because I'm waiting for a riser card for easier measuring and debugging.
I'm still on the logic enable debugging trail. Do you think its safe if I just manually pulled every logic / power input "high"?
With a current limiting resistor and you still have to remove the logic element otherwise it will get cooked. Look for a 5-leg sot23 IC.

Every graphics card that I know has some sort of "no 12V rail connection" display output. So the user knows when he forgot to connect the big +12V power cables.
So 1,8V, 3,3V and 5V around the display connectors should be present for that. I only see the 3,3V.
The first part is just an assumption. I have seen cards that will not start at all.

Also my 5V rail indeed seems to have very low resistance... its only [EDIT!] ~7.3 Ohms. I think thats very low. But I'm not sure. Maybe there IS still an issue with the MOSFETS or driver ICs somewhere.
I would check the resistance with the regulator removed, isolating the supply from the load.
 
Feb 6, 2021
10
1
15
Update: Probably bad news (for my card).

A friend of mine needed to have his thermal paste renewed for his 1080 TI. Luckily he has the MSI 1080 Ti Gaming X from the video above.
So I helped him out and measured some resistance values before and after VRM modules.
I didn't know that memory power supply is completely seperated from the GPU supply. So I could have found that something's wrong earlier.
My Memory power supply is ~ 0,0x Ohms. Thats bad.


(Resistances measured to GND)Intact 1080 TI Gaming XMy Broken 1080 TI Armor
PEX / PLL Output80-100 Ohms100 Ohms
5V Rail (MP1475)Cycling value7 - 15 Ohms
Memory VRM Output80+ Ohms0,0x Ohms
GPU VRM Output0,0x Ohms0,0x Ohms
1,8V (near display port)720 Ohms2,3 Ohms

So I went on and tried to figure out where the short on the memory VRM comes from. I took a IR scanner applied 0,7V at 2A and saw a temp rise of around 10°C at a GDDR5X chip. :( I have to verify it with some ice spray, but its very likely not a simple part. So thats really disappointing, because I have no safe BGA soldering equipment. I have 2 hot air soldering devices tho. But at this point I'm just thinking about selling the defective card. At the moment (BTC at 50k+USD) I almost get the same price for my defective card as for a working card 4 months ago.
 
Last edited:
Update: Probably bad news (for my card).

A friend of mine needed to have his thermal paste renewed for his 1080 TI. Luckily he has the MSI 1080 Ti Gaming X from the video above.
So I helped him out and measured some resistance values before and after VRM modules.
I didn't know that memory power supply is completely seperated from the GPU supply. So I could have found that something's wrong earlier.
My Memory power supply is ~ 0,0x Ohms. Thats bad.


(Resistances measured to GND)Intact 1080 TI Gaming XMy Broken 1080 TI Armor
PEX / PLL Output80-100 Ohms100 Ohms
5V Rail (MP1475)Cycling value7 - 15 Ohms
Memory VRM Output80+ Ohms0,0x Ohms
GPU VRM Output0,0x Ohms0,0x Ohms
1,8V (near display port)720 Ohms2,3 Ohms

So I went on and tried to figure out where the short on the memory VRM comes from. I took a IR scanner applied 0,7V at 2A and saw a temp rise of around 10°C at a GDDR5X chip. :( I have to verify it with some ice spray, but its very likely not a simple part. So thats really disappointing, because I have no safe BGA soldering equipment. I have 2 hot air soldering devices tho. But at this point I'm just thinking about selling the defective card. At the moment (BTC at 50k+USD) I almost get the same price for my defective card as for a working card 4 months ago.

I would take the supply coils off separating the supply and the load and then make measurements...
If the chip input reads 0 Ohm to ground, the card is of no use but parts - getting a good chip is expensive and you need a good IR to install it. But first, remove all VRM coils to find where is the short, (PSU or load).
The IR scanner is very handy tool for finding things, you can easily trace what prevents your enable from being pulled up or down, I would use that to find stuff.
The memory ICs are going bad often (thanks to overclocking) but they are small enough for the heat gun, just preheat the card (with proper profile) to ~150C and you should be fine. Without IR station re-balling would be a pain, so look for ones with balls when you shop.
 
Solution
Feb 6, 2021
10
1
15
To conclude this thread: I sold the card. I stick with my GTX 745 and share a card with a co-worker if I need power for 1-2 days. Yes that is possible if you are careful. :D Until the prices drop some day. I sold the defective card for 175€ with all the problems described. If i just wrote "defective" it would probably have been 250€+

Thanks for all the support vov4ik_il. Much appreciated. And maybe I'll start off with a less expensive card next time :ROFLMAO:
I marked your last answer as best because it would have been the right step to remove the coils first.
 
  • Like
Reactions: vov4ik_il