Rtx 3070 vram extreme throttling

[Devign]

Hello all,

 

A week ago I bought a second hand gigabyte rtx 3070 vision oc. It was running quite hot, about 82-83 degrees and thermal throttling a little bit. So I decided to replace the thermal paste, and because some of the thermal pads where damaged I replaced those too. 

 

I also gave it a slight undervolt, and it ran great. Not going above 65 degrees. Did several 3dmark benchmarks and stress tests, and it ran great for several hours. 

 

Since then I did some moderate gaming for 3-4 days without any issues. Today all of a sudden, after playing rust for 2 hours the card is massively throttling on the vram, lowering from 7800 to 800mhz for some time, then shortly back up to 5000 or 7800 and back to 800 again.

 

I have replaced the vram thermal pads again, and they all seem to make good contact. The throttling is still going on however, almost instantly after I boot up the pc now.

 

Does anyone have a clue what might be the problem? Is my vram fried?

 

And if so, does anyone have an idea if I can send the card to gigabyte for warranty/repair since I got it second hand and replaced the thermal pads?

 

Thanks.

 

Edit: forgot to mention, the 3070 doesn't have a temperature sensor for vram since it's gddr6, so I have no idea whether it's overheating..

[Devign]

Okay small update on my issue, I think I have solved it.

 

The problem was that the vram thermal pad thickness was supposed to be .75mm. I could not find this thickness really, so I bought .5mm and 1mm gelid extreme, as they were supposed to be both relatively soft and high quality.

 

I first tried the 1mm, but got instant 80 hotspot on idle, so no contact with the die. I then tried the .5mm pads, which seemed to work well (for a few days at least).

I'm still not sure how or why it suddenly stopped working properly, maybe the pads got softer because of the sustained heat and compressed to much?

 

I have now put new 1mm pads again, but cut individual squares for each vram chip instead of bigger strips, and I pre-compressed them quite a bit before sticking them on the vram.

This seems to work well enough although the die contact is slightly worse than with the .5mm pads (+15 hotspot instead of +10). I did apply plenty thermal paste and spread it evenly with the small spatula.

 

The instant vram throttling when starting up was some sort of safe mode because of the previous overheating I presume, since it went away after +-2 minutes. Phew.. it really did scare me causing me to instantly turn the pc back off 4 times and getting paranoid I destroyed my card.

 

For the past 20 minutes I have been running the Heaven benchmark, and vram clocks have been steady since. The card max temp is 66.1 with 81.0 hotspot, I suppose these are very acceptable numbers. 

Edited March 31, 2023 by Devign