My reference 7900XT temperature problem

[lixlax]

Hello!

So I bought a reference 7900XT shortly after launch in December. I was so happy seeing the temperatures maxing out at ~65c and ~80c for the junction, excellent cooling performance for such a small card/cooler at 308W.

After 2 weeks or so I noticed that the max junction temps had slowly started climbing and after a few months reaching close to 100c. I thought they probably went out and saved couple of cents by using some terrible thermal paste on a 1000€ card.
At the beginning of the june I finally decided to take it apart to repaste and bought some MX-6. The old paste was quite dry between the die and the cooler, so I thought that could've been the problem indeed.

After repasting the junction temperatures were back to 80c max so I thought I've solved the issue. But after a week I noticed that the max temperatures are slowly climbing again, now barely 1,5 months later the max temps are getting close to 100c again.

Hereby I want to ask the forum members if any of you have encountered a similar problem and how did you fix it? If not then just ideas are welcome.
Right now I'm thinking if too low mounting pressure could be the problem (that after some time the card being mounted the pressure slowly drops due to gravity/sag), but don't have any means to test it. Clearly the vapor chamber can't be the problem.

 

[Shmee]

While I don't know the cause of the temperature increase, you could always try underclocking, undervolting, or reducing power limit. This can help.

 

[Tech Junky]

When I played around with mining I played with pasting as well. Running dual cards in the case though resulted in a heat soak situation. Using a manual fan profile to set the fans at 70% kept things more reasonable.

Since then though I've switched to graphite pads as paste is unreliable for long term use. The pad keeps my laptop 3060 under 70C most of the time. I use the pads though on both GPU and CPU.

Another thing that helped was wedging a fan above the GPUs to blow more air between them inside the case.

 

[WelshBloke]

TBH as long as you are not getting any problems I'd say that junction temperature is normal.

 

[mikeymikec]

The card manufacturer ought to have the best idea of what to expect from the card.

 

[DAPUNISHER]

The card manufacturer ought to have the best idea of what to expect from the card.

That's where I'd begin. If they acknowledge an issue, they will provide an RMA.

It is odd that the temperature keeps creeping up over time. It does lend to the hypothesis there is some physical separation occurring.

 

[lixlax]

I know that up to 110c is "within spec", but it just bothers me.
I actually usually run the card at -10% powerlimit, clock cap at 2400MHz and 1065mV, but even then it quite easily reaches mid to high 80's now. Also I haven't changed the case nor airflow (much), when I take the sidepanel off the temperature drops by 3-4c which is not much I think.

Right now I think I'll run it a little more like this (and see if it the temperature still keeps rising) and then take a look. A graphite pad could be interesting to try since its properties shouldn't degrade at all.

I've never had anything like this with my previous cards- I've only repasted them after ~2 years as a precaution or before resell.

 

[Tech Junky]

I take the sidepanel off the temperature drops by 3-4c which is not much

I tried that approach as well but, wedging a fan between the panel / cards forcing more air between them did more to drop temps.

Hitting those higher temps puts more stress on the solder which will weaken with bigger swings between temps load/idle. I prefer a more stable temp than ultra cool temps. Theoretically though 100C shouldn't impact the solder though.

Erring on the side of caution the lowest temp to melt things would be 90C but, the highest point being 450C and average ~185C would indicate the potential for a shotty solder to come apart at the high temp watermark mentioned @ 110C. So, not knowing which method or metal/alloy being used with these devices makes it hard to know at which point there could be a failure or if we're hitting close to the melting point to cause issues after repeated heating sequences. If the excerpt is correct in that most devices are using a melting point closer to 185C then there's room for overheating with less chance of degradation.

In mining applications though the repeated flex of cards on heat cycles though have shown to be enough to kill cards.

 

[lixlax]

An update on the matter.

Since I made the thread, max temps stopped climbing and never reached 110c when the card starts throttling. But the high temps still bothered me.
I did some more search for similar issues and found out that modern GPUs tend to pump out the thermal paste from between the die and cooler. Reddit particularly seemed to be hyped out about Honeywell PTM7950 phase change thermal material since its solid when cool and turns into paste only when heated up.

So I decided to give it a try ordered a sheet from China and it arrived about a month ago. I surprisingly managed to apply to my GPU on the first try. After installation the max hot spot temps were even 2-3c better compared to freshly applied thermal paste. Now about 1 month later I'm happy to report that the hot spot temperature is still the same and hasn't risen a bit.
The PTM7950 seems like a real deal so far.