Graphics Card Necromancy: Resurrecting a Dead GTX 690

[RazorWind]

Thanks for the reply I really appreciate it.
I am indeed planning on doing a dual 690 setup once I get two cards running stable. It may not be practical but I'm more interested in cool hardware than crazy fps. And at only $89 a card for what was once a $1000 videocard that can still hold its own, I find that awesome.
I bought these cards three weeks ago and was constantly getting dpc watchdog violations and black screens. So after I got the dpc violations figured out the next thing I did was try a different power supply.
My original is a Thermaltake 875watt unit. I ended up taking my EVGA 1300watt out of a computer that I have running dual gtx 590's and it didn't seem to make a difference at all.
I am using an EVGA x99 classified board so I decided to plug in the extra power connector that feeds the pci-e ports...but it didn't make a difference.
I cleaned the pci-e slot but still no joy.
It really wasn't until I received the replacement card Monday that I decided to completely rule out any problems with my computer hardware or software, because the new card works perfectly.
I have a volt meter.... somewhere. I'll hunt for it and post my findings.

Yeah, you still need to do the voltage test. It doesn't matter what the rated wattage of the power supply is - it's the voltage that it actually provides that matters, and that can degrade over time.

Failing that, assuming it's not something like the GPU dies overheating (I assume you checked for this already, beyond just replacing the grease), I'd be inclined to suspect a board issue. It could be a cracked solder ball or trace, degraded dies, or something power related. You could conceivably try turning off SLI to see if the problem goes away when you only run one of the GPUs.

Edit: One other thing to check for is missing tiny SMD capacitors. These are super easy to knock off the board via rough handling, particularly when you see a pile of the cards stacked up like some ebay sellers like to show in their photos. As I showed in my thread about the 290X, you can get this kind of behavior if even one of those tiny caps is missing, if it's the right one.

 

[Dizzydan02]

I checked resistances on all of the pins. I also checked voltages on 3 of the 12v pins, the pin on the bottom row furthest to the left, the pin on the bottom row furthest to the right and the pin on the bottom row second to the last on the right.
Bottom row furthest to the right showed 0 voltage....but when i tested it for ohm's it showed close to 3k. The other two pins showed 11.5v but did not fluctuate at all even when the screen went black. I figured this was low so i decided to put it in the pc with the 1300w psu (which is new, just purchased about a month ago). This showed the exact same results. No voltage at the bottom right pin, constant 11.5v no fluctuations. Maybe my volt meter is reading a little on the low side.
How do i check to see if the gpu dies are overheating, i only replaced the paste and pads. I will do a side by side comparison and see if there are any components missing.
Another thing to mention that I had forgot about. Both cards would black screen just sitting at the desktop not doing anything, that is until I adjusted a setting in the nvidia control panel. I set the power management mode from optimal to performance. Which raised the static voltage from .987 to 1.050
Just this change prevented the computer from black screening at the desktop.


I only uploaded pictures for the left bank, the right bank gives the exact same readings. The second to last picture is from the pc with the 875w psu. The last pictures is from the pc with the 1300w psu.

 

[RazorWind]

Your resistance measurements look good. A few thousand is what you should expect to see across a whole graphics card.

11.5 volts is way too low. You need twelve. actual. volts. It may just be that your voltmeter reads a hair low, but if that's accurate, that's your problem right there. I'd try to get my hands on a digital one and do the voltage test again. You should definitely see a small difference in voltage between idle and running a game - like .2 volts or less. In the pictures, is the card running a game, or just idling?

You can use GPU-Z or Afterburner to check the GPU die temperatures, but that's probably not the problem if it's not stable even at idle. I think I might check the actual core voltage with a digital multimeter next, to see if it agrees with what's reported by GPU-Z or afterburner. If it's lower than expected, that will lead to instability as well.

 

[Burticus]

This is nuts and awesome at the same time. Thumbs up

 

[Dizzydan02]

Both pictures showing volt readings were at idle. But there was absolutely no change when I started a game. No fluctuations at idle, during game or when the black screen happened.
Kind of a weird situation because all of these cards are new and the replacement one actually works perfect.
I have monitored temperatures ever since i got these cards and they have been normal i guess i just didnt understand the question.
Ill do some searching online for a good digital meter and redo the tests.

 

[RazorWind]

Wha, they're new, and not refurbs? Were they some sort of new old stock? Did you get boxes with them? If so, that's pretty cool. I have four (three dead, one I repaired), and I'm pretty at least two were used for mining. All four were clearly... experienced, though.

 

[Dizzydan02]

Well they were advertised as used but they all look new. No dust, marks, scuffs or scratches anywhere on them. Even still have the original protective plastic on the clear plastic heatsink covers.
I thought maybe someone could have took the cards apart after purchase and put water blocks on the boards. However the boards look as new as the heat sink fan and cover assembly and they all had the original tim, paste and pads.

 

[Dizzydan02]

I think I found the problem. Look closely as only a trained eye can spot such defects.

 

[RazorWind]

I think I found the problem. Look closely as only a trained eye can spot such defects.
View attachment 219758

I can't tell if you're just being snarky about the "Made in China" sticker or there's actually something wrong, but I don't see any visible damage.

 

[Dizzydan02]

Yeah i was just goofin. I should have my digital meter by Monday and that company is going to seen me another card. Hopefully I'll get that by Wednesday.

 

[Dizzydan02]

I did get the new meter in and tested the voltage to the card and it holds 12.22v at idle and goes down to 12.21v when the screen loses signal and the fan spins up to 100%.
I also applied pads to the mosfets that were not originally padded like you had done in a previous posting and I am able to actually play a game.
Resident Evil Revelations is the game I was testing with, which uses both gpu's. Afterburner and Rivaturner are the programs I was also using to monitor temps, clock speeds and voltages.
I could play for forever while the cards voltage was at the stock .987 but once it started to fluctuate to 1.012 and 1.025 it only took approximately 15 seconds and it did the ol' no signal & 100% fan trick.
But yeah, adding pads to those mosfets definetly helped. I couldnt even get into a game before with this particular card.
And this was tested in the pc with the 875w power supply and i tested on both 8 pin connectors.

 

[RazorWind]

I did get the new meter in and tested the voltage to the card and it holds 12.22v at idle and goes down to 12.21v when the screen loses signal and the fan spins up to 100%.
I also applied pads to the mosfets that were not originally padded like you had done in a previous posting and I am able to actually play a game.
Resident Evil Revelations is the game I was testing with, which uses both gpu's. Afterburner and Rivaturner are the programs I was also using to monitor temps, clock speeds and voltages.
I could play for forever while the cards voltage was at the stock .987 but once it started to fluctuate to 1.012 and 1.025 it only took approximately 15 seconds and it did the ol' no signal & 100% fan trick.
But yeah, adding pads to those mosfets definetly helped. I couldnt even get into a game before with this particular card.
And this was tested in the pc with the 875w power supply and i tested on both 8 pin connectors.
View attachment 222236

Are the MOSFETs you're referring to Q15 and Q16? It could be that the bootstrap capacitor on that rail isn't working, which would lead to the gate voltage still being low, which is consistent with what you're describing.

 

[Dizzydan02]

Yes I did 15 - 20 and then two memory modules.
I borrowed a pic of yours to illustrate.

 

[RazorWind]

Yes I did 15 - 20 and then two memory modules.
I borrowed a pic of yours to illustrate.
View attachment 222243

The components you have circled in orange are the primary VRM controllers. As I recall, they should have a thermal pad from the factory. The components circled in blue near the center of the picture are the MOSFETs for the .95V rail. You shouldn't have to put a pad on them - when they overheat, you generally get that zero ohm resistor next to them burning out before they do, and they don't have any thermal monitoring to actually shut them off.

The main Vcore VRMs do have thermal protection, though. Are you sure those have the right thickness thermal pads?

 

[Wooshoo]

I build jets for a living, but could never get the EE part down pat. Enjoying this thread immensely.

I'm an electrical engineer for a living and I wish I was this organized and had enough of an idea of how the cards work to not literally go through every single component and look up a corresponding data sheet. This thread is awesome amd I am saving it.

 

[Wooshoo]

This thread made me want to drag out a couple dead 7870s and try to get one working just for kicks. If only there was more time in a day!

 

[RazorWind]

This thread made me want to drag out a couple dead 7870s and try to get one working just for kicks. If only there was more time in a day!

I have a few other similar threads where I've worked on some other cards, including a 290X (which is similar). Would you be interested in selling me those dead 7870s (edit: So I can share working on them here, obviously)?

The data sheets are usually pretty easy to find for cards as old as the 690, although you sometimes only get a vague pinout, and very little explanation of what anything does. On newer stuff, particularly the 10 and 20 series, some vendors seem to have come up with their own bespoke parts, and made the datasheets confidential, not unlike Apple and their ISL9240 that Rossmann likes to kvetch about.

 

[Wooshoo]

I have a few other similar threads where I've worked on some other cards, including a 290X (which is similar). Would you be interested in selling me those dead 7870s (edit: So I can share working on them here, obviously)?

The data sheets are usually pretty easy to find for cards as old as the 690, although you sometimes only get a vague pinout, and very little explanation of what anything does. On newer stuff, particularly the 10 and 20 series, some vendors seem to have come up with their own bespoke parts, and made the datasheets confidential, not unlike Apple and their ISL9240 that Rossmann likes to kvetch about.

I dont know! Honestly if you had asked me before I saw your thread I would have said yes immediately but now I'm thinking about tearing into them. I'll let you know if I can convince myself to let them go

 

[talmas]

Hi RazorWind! I'm sorry to invade your post, but I think you could help me. I got a gtx 690, and I accidentally broke a component (U 507), which is the one in the red circle in the images. I think the component is a current monitor, specifically A219. Can you confirm what this component is by looking at your GPU?

 

[RazorWind]

Hi RazorWind! I'm sorry to invade your post, but I think you could help me. I got a gtx 690, and I accidentally broke a component (U 507), which is the one in the red circle in the images. I think the component is a current monitor, specifically A219. Can you confirm what this component is by looking at your GPU?

View attachment 226045View attachment 226044View attachment 226043

The markings on U507 on my cards say "A219." Some cursory googling indicates you're probably right, that it's a Texas Instruments INA219.
http://www.ti.com/lit/ds/symlink/ina219.pdf

The real problem now becomes how to attach it, given that it appears that six of the pads got ripped off the board. The more instances I see where failures happen because something got scraped off the back like this, the more convinced I am that this is the real reason new graphics cards so often come with a metal shield on the back.

 

[talmas]

The markings on U507 on my cards say "A219." Some cursory googling indicates you're probably right, that it's a Texas Instruments INA219.
http://www.ti.com/lit/ds/symlink/ina219.pdf

The real problem now becomes how to attach it, given that it appears that six of the pads got ripped off the board. The more instances I see where failures happen because something got scraped off the back like this, the more convinced I am that this is the real reason new graphics cards so often come with a metal shield on the back.

Thank you! I'll order some INA219 from texas instruments and try to fix my 690.

And yes, if my GPU had a backplate, this wouldn't have happened

 

[mrExodus]

Hi RazorWind, sorry for necroing your necromancy thread but I would like to ask for some help with my used GTX 690. It seems someones screwdriver had a fight with it and now is missing some resistors, also the seller though a wallmart bag and a small box would be enough packaging so part of the shroud is broken and the GTX logo is kinda crushed

 

[RazorWind]

How can I help you? Do you need me to just tell you the values for the missing components so you can replace them, or did you need me to actually repair this card for you?

 

[mrExodus]

The value of the knocked resistors would be great; I have no issue soldering SMD components

 

[RazorWind]

First picture, near C785: 10Kohm

Second picture, near Q515: 5Kohm

Third picture, near U504: Both are 10 ohms

Lucky for you, I have a literal pile of dead 690s...

 

[Mode13]

Razor, after going over my own board (the 780Ti mentioned in the classified thread), I've kind of come to figure that the black top SMDs are resistors and the light brown, would that be true?

 

[RazorWind]

Razor, after going over my own board (the 780Ti mentioned in the classified thread), I've kind of come to figure that the black top SMDs are resistors and the light brown, would that be true?

The black components are usually resistors. The tan components are usually capacitors of the multilayer ceramic type.

 

[Mode13]

Thanks, yeah kind of tired here and didn't complete my own post.. light brown / tan being caps.

I wont bother asking if there's a way to guess the capacitance / operating voltage on the caps since there's no markings at all.. Techpowerup does have high res shots of the PCB, so at least i can figure out if i need a cap or resistor then. I'm assuming my only choice would be to remove caps from the donor board and test them with my meter..

I'm assuming that's what's caused the failure on the card.. The business side of the card survived transit due to having the heatsink on, but the back with no backplate clearly got rubbed against and took significant damage.. I know you can survive with 1 or 2 little noise filtering caps falling off.. but 10+..? sigh

 

[RazorWind]

Thanks, yeah kind of tired here and didn't complete my own post.. light brown / tan being caps.

I wont bother asking if there's a way to guess the capacitance / operating voltage on the caps since there's no markings at all.. Techpowerup does have high res shots of the PCB, so at least i can figure out if i need a cap or resistor then. I'm assuming my only choice would be to remove caps from the donor board and test them with my meter..

I'm assuming that's what's caused the failure on the card.. The business side of the card survived transit due to having the heatsink on, but the back with no backplate clearly got rubbed against and took significant damage.. I know you can survive with 1 or 2 little noise filtering caps falling off.. but 10+..? sigh

It depends on exactly where the components are. There are some places you'll see them used that have a specific value, per some published specification. For instance, if you look at the traces near the PCI-E connector, you'll notice there's a row of tiny caps that are in line with the traces, two for each lane. Those have a known value (100nF, I think?). Other functions also tend to have predicable values. 10K resistors are pretty common in pull-up aplications. 10 ohm resistors are common in pull down applications, and so forth.

But, even when you think you know a component's value, if at all possible, you really do need to actually check before you solder a new one into your circuit.

 

[dragrdan]

I expected to see a 690 in an Easy Bake Oven! Jk, this is the kind of stuff I love to read about. Hope you keep posting updates, regardless of the results!

Can a defective 690 still function as a 680? Depending on the damage of course.

I have a 690 that one gpu isn't working. The system sees it as a single 680