6800xt is dying (idk why but I found something interesting)

[Full_Logicz]

Story time:

Last week my pc started to randomly crash to black screen, no BSOD, driver crash or other usual crashes and I started investigating and I did a bunch of fixes that worked for other people in forums and such, fast forward to today I know 100% the GPU is the cause of the crashes (tried the GPU in 2 pcs different drivers, wall plug, HDMI DP cables, monitors, USB devices literally anything that would be a constant with a crash happening). So I decided to do some logs of the GPU with HWinfo64 of the GPU before crashing maybe I could learn something from that before sending the GPU to warranty. After I took the GPU out I found something that looked like oil stains on the back of the PCB and I'm so confused.

 

And I see some very strange frequency/voltage interactions on the logged graphs which I attached the photo below for. (I used 3dMark for the benchmark Firestrike GPU scenes on loop)

 

Questions:

 

Is this graph normal operation for Radeon 6000 series ?

Might this be the cause of the random black screens I was getting ?

Can anyone teach me something based on my findings ?

 

What are the oil stains from the PCB ?

Where are they from ? (I don't remember spilling oil on the GPU)

 

Happy gaming everyone !

 

Left graphs: 1080ti     Right graphs: 6800xt

[191x7]

6 minutes ago, Full_Logicz said:

Story time:

Last week my pc started to randomly crash to black screen, no BSOD, driver crash or other usual crashes and I started investigating and I did a bunch of fixes that worked for other people in forums and such, fast forward to today I know 100% the GPU is the cause of the crashes (tried the GPU in 2 pcs different drivers, wall plug, HDMI DP cables, monitors, USB devices literally anything that would be a constant with a crash happening). So I decided to do some logs of the GPU with HWinfo64 of the GPU before crashing maybe I could learn something from that before sending the GPU to warranty. After I took the GPU out I found something that looked like oil stains on the back of the PCB and I'm so confused.

 

And I see some very strange frequency/voltage interactions on the logged graphs which I attached the photo below for. (I used 3dMark for the benchmark Firestrike GPU scenes on loop)

 

Questions:

 

Is this graph normal operation for Radeon 6000 series ?

Might this be the cause of the random black screens I was getting ?

Can anyone teach me something based on my findings ?

 

What are the oil stains from the PCB ?

Where are they from ? (I don't remember spilling oil on the GPU)

 

Happy gaming everyone !

 

Left graphs: 1080ti     Right graphs: 6800xt

 

 

 

What's the rest of the specs?

 

The oily stains are (highly likely) silicone oil from the thermal pads.

[IkeaGnome]

What were the temps? This almost looks like it's trying to thermal throttle or it's not under a full load at a few points. What game or benchmark did you run to get these graphs?

That oil is more than likely from the thermal pads. They've got an oil in them that tends to leak over time. It's not conductive and isn't anything to really worry about. 

[Full_Logicz]

1 minute ago, 191x7 said:

What's the rest of the specs?

 

The oily stains are (highly likely) silicone oil from the thermal pads.

Ryzen 9 5900x

Arctic freezer 280mm

ASUS TUF GAMING X570-PLUS

4x8 G.Skill 3600mhz cl16

xfx merc 319 black 6800xt 

FSP Aurum 850w 80+ plat

 

Currently using a 1080ti 

 

 

[Full_Logicz]

4 minutes ago, IkeaGnome said:

What were the temps? This almost looks like it's trying to thermal throttle or it's not under a full load at a few points. What game or benchmark did you run to get these graphs?

That oil is more than likely from the thermal pads. They've got an oil in them that tends to leak over time. It's not conductive and isn't anything to really worry about. 

Temps are 75 average 90 hotspot, HWinfo made a graph of throttle for current temperature and power and it never hits a limit in the 20 minute log

[IkeaGnome]

1 minute ago, Full_Logicz said:

Temps are 75 average 90 hotspot, HWinfo made a graph of throttle for current temperature and power and it never hits a limit in the 20 minute log

Is HWinfo showing any reason for the graphics card to throttle? HWinfo basically only shows thermal throttling, GPU-Z will show more. Either of them will tell you this in their sensors mode. HWinfo is a tab, GPU-Z is a check box when you run the application.

https://www.techpowerup.com/gpuz/

[Full_Logicz]

4 minutes ago, IkeaGnome said:

Is HWinfo showing any reason for the graphics card to throttle? HWinfo basically only shows thermal throttling, GPU-Z will show more. Either of them will tell you this in their sensors mode. HWinfo is a tab, GPU-Z is a check box when you run the application.

https://www.techpowerup.com/gpuz/

I used HWinfo because of the faster update rate on logging 20ms vs 100ms on gpuz

[IkeaGnome]

Those temp graphs scream of it throttling. 

20ms vs 100ms isn't going to make a difference in logging here. Really, we don't even need to log. It'll still be stored in the GUI's min/max/current sections. 

GPU-z's GUI shows more in a condensed place as well. If you want to keep to HWinfo, we can, but there's a whole lot more information in the "sensors only" mode shown above. On your graphs, there is a throttle at 11:30, but no hints as to what it's for. 

 

What application are you using to put the graphics card under a load when you get those graphs.

[Full_Logicz]

2 minutes ago, IkeaGnome said:

What application are you using to put the graphics card under a load when you get those graphs.

It's Timespy on loop because i'm tired of waiting for a game to crash. The problem i see in the original picture i posted is that the gpu is trying to go to a higher frequency from idle and the voltage dunks down to 0v that is the question for me mostly. The temps on this gpu have been like this since i bought the thing so i'm not especially worried about the temps since it's been running fine up until a week ago when it started crashing roughly every hour or so, no event viewer log nothing. i tried multiple drivers i took it out there was no dust on it or in it i checked with a flashlight, tried 3 drivers(newest 22.something and the launch driver) put it in 2 different computers to test both did the same I literally tried everything (even win11 vs win10 because the other computer is a bit older and i couldn't install 11 on it)

[IkeaGnome]

7 minutes ago, Full_Logicz said:

It's Timespy on loop because i'm tired of waiting for a game to crash. The problem i see in the original picture i posted is that the gpu is trying to go to a higher frequency from idle and the voltage dunks down to 0v that is the question for me mostly. The temps on this gpu have been like this since i bought the thing so i'm not especially worried about the temps since it's been running fine up until a week ago when it started crashing roughly every hour or so, no event viewer log nothing. i tried multiple drivers i took it out there was no dust on it or in it i checked with a flashlight, tried 3 drivers(newest 22.something and the launch driver) put it in 2 different computers to test both did the same I literally tried everything (even win11 vs win10 because the other computer is a bit older and i couldn't install 11 on it)

Can we see the Sensors, or get the .csv log file from Hwinfo? Either of them will get a whole lot more information than a couple small graphs at a time. The .csv will take the longest to go through, but will show sensors for everything in your computer and all the points that it recorded.

[Full_Logicz]

5 minutes ago, IkeaGnome said:

Can we see the Sensors, or get the .csv log file from Hwinfo? Either of them will get a whole lot more information than a couple small graphs at a time. The .csv will take the longest to go through, but will show sensors for everything in your computer and all the points that it recorded.

This is the longest log with all the gpu data, i excluded cpu mobo and all the other stuff since it just bloated the log. everything looks roughly like what i see online in other graphs about the 6800xt besides those voltage drops to 0V. Currently the gpu is packed for warranty which i will send tomorrow.

 

This isn't actually the first log it's the 3rd because i made the first 2 txt by mistake but i named it 1 since it's the first logged properly.

CrashHw1.CSV

[IkeaGnome]

33 minutes ago, Full_Logicz said:

This is the longest log with all the gpu data, i excluded cpu mobo and all the other stuff since it just bloated the log. everything looks roughly like what i see online in other graphs about the 6800xt besides those voltage drops to 0V. Currently the gpu is packed for warranty which i will send tomorrow.

 

This isn't actually the first log it's the 3rd because i made the first 2 txt by mistake but i named it 1 since it's the first logged properly.

CrashHw1.CSV 14.7 MB · 0 downloads

That looks a whole lot like power delivery issues on the graphics card. VRMs, Mosfets etc. If it helps, it looks like power delivery to the vram is still working fine. It's not throttling, it's not self lowering PPT limit, it just looses power to the GPU core. If you look at line 3330, voltage and watts to the GPU just drop out. 

[Full_Logicz]

6 minutes ago, IkeaGnome said:

That looks a whole lot like power delivery issues on the graphics card. VRMs, Mosfets etc. If it helps, it looks like power delivery to the vram is still working fine. It's not throttling, it's not self lowering PPT limit, it just looses power to the GPU core. If you look at line 3330, voltage and watts to the GPU just drop out. 

interesting but it does this in two completly different systems my current system with the 850w FPS psu and my old system with a 750W seasonic so might the gpu power delivery be faulty ? the gpu is about 1.5 years old at this point.

[IkeaGnome]

2 minutes ago, Full_Logicz said:

interesting but it does this in two completly different systems my current system with the 850w FPS psu and my old system with a 750W seasonic so might the gpu power delivery be faulty ? the gpu is about 1.5 years old at this point.

If you've got dying components in the power delivery system of the GPU it would stay with the GPU in a different system. There are some tests you can do, but they will void your warranty, so I'd just warranty it at this point.

[Full_Logicz]

3 minutes ago, IkeaGnome said:

If you've got dying components in the power delivery system of the GPU it would stay with the GPU in a different system. There are some tests you can do, but they will void your warranty, so I'd just warranty it at this point.

Tests as in running the gpu some stress tests or opening it up and tinkering a bit ? If i mark the last message of yours as a solution could you still inform me of the tests ?

[IkeaGnome]

Just now, Full_Logicz said:

Tests as in running the gpu some stress tests or opening it up and tinkering a bit ? If i mark the last message of yours as a solution could you still inform me of the tests ?

Opening up and getting at it with a multimeter. I know you want to learn diagnosing and what not, but I'd be a bit hard pressed to run you through schematics and testing on a gpu that could be warrantied. If it wasn't, presumably, your main graphics card then it'd be a different story.

 

Out of curiosity, which 6800 xt is it?

[Full_Logicz]

3 minutes ago, IkeaGnome said:

Opening up and getting at it with a multimeter. I know you want to learn diagnosing and what not, but I'd be a bit hard pressed to run you through schematics and testing on a gpu that could be warrantied. If it wasn't, presumably, your main graphics card then it'd be a different story.

 

Out of curiosity, which 6800 xt is it?

Yeah i'd love to give that a go but i'm short on money and i'd rather warranty it. 

It's this RX-68XTACBD9: https://www.xfxforce.com/shop/xfx-amd-radeon-tm-rx-6800xt-16gb-gddr6-merc-319

[IkeaGnome]

5 minutes ago, Full_Logicz said:

Yeah i'd love to give that a go but i'm short on money and i'd rather warranty it. 

It's this RX-68XTACBD9: https://www.xfxforce.com/shop/xfx-amd-radeon-tm-rx-6800xt-16gb-gddr6-merc-319

That is a reference design board and I've got schematics for it. Just to give you a bit of a clue where I'd start. Since it's GPU only and not also Vram, I'm not worried about any circuits they both share. The first major set of circuits they don't share is the power delivery. Red circle is for the memory, and blue is for the GPU.

I'd start by probing to look for anything with a short to ground to be safe, but not expect to see one. Then check the inductors and mosfets for odd readings. The inductors are the ones marked "R15 2031". The Mosfets are the small square IC's just to the outside of the inductors. 

https://www.infineon.com/cms/en/product/power/dc-dc-converters/integrated-power-stages/tda21472/

And here is how that would look in a circuit diagram for this board.

 

It's hard to walk you through it without actually doing it, but I figured since you mentioned wanting to learn a bit in your OP I figured I'd at least get you my first couple steps. 

 

Something to note is that these MOSFETs drop off in delivery performance with temperature.

 

With what your GPU is doing I'd probably rig up a cooler to not fry the GPU and power it on for a bit to let the Mosfets get nice and warm and see if there's one that has drastically different readings from the rest or when they're all cool. 

[Full_Logicz]

4 minutes ago, IkeaGnome said:

 

With what your GPU is doing I'd probably rig up a cooler to not fry the GPU and power it on for a bit to let the Mosfets get nice and warm and see if there's one that has drastically different readings from the rest or when they're all cool. 

thanks a lot for taking time to give detailed responses, depending on what happens with the warranty I might get back a gpu that will fail in the future and i can play around with it, cheers and have a good one!

[IkeaGnome]

1 minute ago, Full_Logicz said:

thanks a lot for taking time to give detailed responses, depending on what happens with the warranty I might get back a gpu that will fail in the future and i can play around with it, cheers and have a good one!

If troubleshooting something like that is something you want to learn, and you have a multimeter just keep an eye out for a broken card or two and feel free to message me here. Schematics are a bit hard to find since they're not meant to be released but I've got quite a few of them and could spend some time helping you with them. With some asking around locally you might be able to get one for free even. No guarantees they're all fixable, but it at least gets some learning in. 

[Full_Logicz]

8 hours ago, IkeaGnome said:

If troubleshooting something like that is something you want to learn, and you have a multimeter just keep an eye out for a broken card or two and feel free to message me here. Schematics are a bit hard to find since they're not meant to be released but I've got quite a few of them and could spend some time helping you with them. With some asking around locally you might be able to get one for free even. No guarantees they're all fixable, but it at least gets some learning in. 

At some point this summer I'll probably give it a go.