[SOLVED] Worth replacing thermal pads on 3080FE?

[_dawn_chorus_]

I was reading this can lower temps dramatically but my question is: does it matter for me?
Had my 3080FE just shy of a year and the temps seem ok and are well "within spec" but if it could be significantly quieter that would be awesome. It sounds like most people doing this mod are mining though and getting thermal throttled by like 110c vram. Since I only game and temps aren't super high I wasn't sure it would be worth it.

After running Deathloop maxed out fps unlocked for 40 minutes my temps read from HWinfo where max:
GPU core - 72.5c
Memory Junction - 92c
Hotspot Temp - 82.7c
Fans at 71%

Deathloop allocates like 96% of my vram and maxes out my gpu so I thought it was a good choice. Those temps will all raise 4c on the nose with fans at 50%.
Anyway it would be awesome to get those same temps or so with fans at say 50% max for a quieter experience.
Does that seem possible with thermal pad replacement or do the returns diminish the lower your temps are to begin with?

 

[Phaaze88]

https://www.micron.com/products/ultra-bandwidth-solutions/gddr6x
You are still within a safe range(below 95C).
Replacing thermal pads is not as simple as some have made it out to be.
-Pads between different manufacturers have not been tested in the same conditions. You know how for cpu cooler thermal tests, there's those dummy heaters? There isn't anything like that done with these pads.
So the behavior you get out of different pads with similar specs could be radically different.

-It's not enough to know pad thickness, there's also texture(very soft, soft, medium, firm, very firm).
If you don't match the originals to a T, you can end up with worse performance, such as the pads having too much/not enough give when securing the cooler on the PCB.

How do you find out? From people who've already tried replacing pads for your specific model.


Undervolting the gpu might help. Undervolting doesn't directly affect the Vram, because the voltage going to it is at a fixed value from vbios, but you may be able to reduce it due to the gpu radiating a little less heat overall.
Run Msi Afterburner. Unlink the Power and Temperature Limits - there's a paperclip like icon next to it depending on the skin being used.
Max out only the power limit, and click apply.
Play your games with Afterburner's own hardware monitor running. Make sure both Core Clock and Gpu Voltage are visible. If not, you can open them by going into Settings > Monitoring tab, and checking them.
After a few minutes, check Afterburner's hardware monitor for the MAX Core Clock and Gpu Voltage. Memorize, or write 'em down.
Close the game, and open Afterburner's Curve Editor. Take the max gpu voltage you recorded, and subtract 0.05v from it. Find the voltage point in the Curve Editor that matches it, or is the closest match, and click on it.
Then use the up arrow key and raise the frequency back up to the max core clock the gpu touched. Lock it with the L key, and click Apply again.
Save the settings in one of the numbered profiles and lock them. Then click the reset key.
Done.

When you want to start up a game, open Afterburner, click the numbered profile it was saved at, and click apply. When you're done, reset it.

 

[_dawn_chorus_]

https://www.micron.com/products/ultra-bandwidth-solutions/gddr6x
You are still within a safe range(below 95C).
Replacing thermal pads is not as simple as some have made it out to be.
-Pads between different manufacturers have not been tested in the same conditions. You know how for cpu cooler thermal tests, there's those dummy heaters? There isn't anything like that done with these pads.
So the behavior you get out of different pads with similar specs could be radically different.

-It's not enough to know pad thickness, there's also texture(very soft, soft, medium, firm, very firm).
If you don't match the originals to a T, you can end up with worse performance, such as the pads having too much/not enough give when securing the cooler on the PCB.

How do you find out? From people who've already tried replacing pads for your specific model.


Undervolting the gpu might help. Undervolting doesn't directly affect the Vram, because the voltage going to it is at a fixed value from vbios, but you may be able to reduce it due to the gpu radiating a little less heat overall.
Run Msi Afterburner. Unlink the Power and Temperature Limits - there's a paperclip like icon next to it depending on the skin being used.
Max out only the power limit, and click apply.
Play your games with Afterburner's own hardware monitor running. Make sure both Core Clock and Gpu Voltage are visible. If not, you can open them by going into Settings > Monitoring tab, and checking them.
After a few minutes, check Afterburner's hardware monitor for the MAX Core Clock and Gpu Voltage. Memorize, or write 'em down.
Close the game, and open Afterburner's Curve Editor. Take the max gpu voltage you recorded, and subtract 0.05v from it. Find the voltage point in the Curve Editor that matches it, or is the closest match, and click on it.
Then use the up arrow key and raise the frequency back up to the max core clock the gpu touched. Lock it with the L key, and click Apply again.
Save the settings in one of the numbered profiles and lock them. Then click the reset key.
Done.

When you want to start up a game, open Afterburner, click the numbered profile it was saved at, and click apply. When you're done, reset it.

So I am not seeing anything for "GPU Voltage" in afterburner. It isn't an option for me in the monitoring tab. I just went off the numbers from hwinfo.

The undervolt definitely made a quieter card! I have about 1-2c less core temp but my fans are down 15% each which was a bigger leap than I expected. As you mentioned though it hasn't helped the Mem Junction temps which now that I am playing closer attention typically seem to hover around 96c but the highest has been 100c.
I wasn't really concerned about them but someone in one of the reddit threads I was reading claimed that GDDR6x degrades quicker and is more sensitive to temperatures than other memory. They did not cite the source and I remain skeptical as I couldn't find anything confirming that. I see on the Micron page you linked though they now mention 95c/105c which seems strange... which one is it?? I mean is 105c Tjmax or not Micron?? lol.

It seems like their are a lot of guides for moding the pads with specific pads recommended based on density and malleability (Gelid Extreme). Even showing cut patterns and thicknesses required. Sounds like the only really hard part is managing the ribbon cables

A certain amount of people do seem to be saying they saw their core go UP a couple degrees despite their mem junc temps being down around 15-20c. I cant seem to find a lot of people talking about core temps though with actual numbers. All the threads I have found are mining related as apparently that cooks the memory, but all these miners also have their cards core underclocked. So its hard to parse apart the info in any meaningful way for my use case.

So really though should I worry about these temps degrading or hindering performance? If I am going to mod the card I would like to do it soon as my EVGA 3080 FTW3 queue position from last October is about to come up, so if I royally F the FE I'll have a back up at least..
.

 

[Phaaze88]

So I am not seeing anything for "GPU Voltage" in afterburner. It isn't an option for me in the monitoring tab.

https://imgur.com/zkv3Nhg

View: https://imgur.com/zkv3Nhg

https://imgur.com/y0qAXwK

View: https://imgur.com/y0qAXwK

I see on the Micron page you linked though they now mention 95c/105c which seems strange... which one is it?? I mean is 105c Tjmax or not Micron?? lol.

All I know is that 95C was there from the beginning and 105C was added later.
Whether it's technically ok or not, I don't believe any of the less technical users would be ok with seeing thermal encroach on 100C, so I'd go with the one that was there first.

A certain amount of people do seem to be saying they saw their core go UP a couple degrees despite their mem junc temps being down around 15-20c.

The cooler probably doesn't mate with the gpu die like it used to. If you've seen Gamers' Nexus images of cooler cold plate pressure mapping, the before and afters with changing thermal pads likely wouldn't be the same.

I cant seem to find a lot of people talking about core temps though with actual numbers. All the threads I have found are mining related as apparently that cooks the memory, but all these miners also have their cards core underclocked. So its hard to parse apart the info in any meaningful way for my use case.

The memory clock can be changed, but nothing else - not even the voltage. It stays the same regardless of the clock speed.
If the core clock is left alone while mining, it wastes a lot of power, thus they underclock and undervolt the core - which indirectly affects the memory since it reduces power consumption across the board.
Since the memory shares the heatsink with the other components, the memory is in turned exposed to less heat.

So really though should I worry about these temps degrading or hindering performance? If I am going to mod the card I would like to do it soon as my EVGA 3080 FTW3 queue position from last October is about to come up, so if I royally F the FE I'll have a back up at least..

If just for gaming, I wouldn't worry about it. I feel that wouldn't fly for mining though.
The performance of the various coolers were tested around games. Mining 'hits differently' from what was intended, and for a number of the models with GDDR6X memory, the cooling doesn't handle it well.

 

[cryoburner]

You might also look into reducing your overall case temps to see if that helps. Having case fans ramp up slightly higher might potentially reduce overall system noise if it allows you to keep the graphics card's fans at a lower speed. Or maybe just rearranging the case fans to alter the path of airflow through the case. Or, if an AIO cooler is in use, switching the radiator from dumping the CPU's heat inside the case to dumping it outside. Doing that would increase CPU temperatures, but could be a viable option if your processor has a lot of thermal headroom to spare.

Replacing the thermal pads of a year-old high-end graphics card seems a bit risky. And if the manufacture where to deem a hardware failure to be the result of the modifications, they could potentially deny a warranty claim.