Overclock.net banner
1 - 20 of 20 Posts

·
Registered
Joined
·
114 Posts
Discussion Starter · #1 · (Edited)
Hello, reposting this from other place as it hasn't got much traction there. Please take a note this is mostly theoretical material, although based on some practical researches. I don't have suitable card to test and confirm reliability and safety of the methods below. But let's get into it anyway xD
Essentialy, gddr6\x overheating case comes down to same problem that forced people to delid intel cpu's. Which is dense heat flow resulting in one side remaning cold while other is overheating. Picture below shows how all the heat output concentrated in small area of 30 mm2:
2480260
So the solution would be to increase heat exchange area and decrease the gap between vram chips and heatsink, which I suggest to use copper shims for:
2480236
But before that I have to warn about all the risks that come into my mind. Since copper shim extends past vram chip, it can short or even crack MLC caps or whatever elements placed around. Another concern, that tinkering with vram gap thickness you can end up with uneven or too high or too low pressure, risking to crack gpu die, bend PCB or disturb thermal contact for some hot elements, including gpu itself. Please watch out for all of this.
There are few approaches to use copper shim:
1) Vram chip > thermal paste > shim > 1mm+ thermal pad > coldplate. I wouldn't use thinner pads as they are not adjustable enough. The softer pad the better, with copper shim increasing heat exchange area by 6 times and reducing gap, you don't need to worry about thermal conductivity much.
2) Vram chip > thermal pad > shim > thermal paste > coldplate. In this case the shim wouldn't work as heatspreader, but still reduce the gap. It's suitable when there are elements around vram chips that would touch the copper otherwise.
3) Vram chip > thermal paste > shim > ~0.5 mm of thermal putty > coldplate. Best solution in terms of thermals, and not only for vram, but also gpu, because all the pressure applied to it exclusively. Which is also a source of some risk. Another concerning point is that shims are not secured in place and can slowly slide off. Which I honestly doubt would happen, but keep possibility in mind.
4) Same, but using a mix of thermal paste and non acid RTV silicone gasket instead of putty. I know this stuff is pure insulator, but I tested it and it's shouldn't be that terrible. Just don't make the gap wider than 0.5mm and use best performing paste you can find. Silicone portion should be in 20-40% range (visually), depends whether you want better thermals or better stability. At 40% it turns into completely dry soft rubber in 12-48 hours.
Here's example of proportions that vulcanized into soft rubber and showed thermal performance around 1w/mk in my testing:
2480259

PS Shims may often require a bit of sanding to make them flat and smooth.
PPS Do not use thermal pads and putty together, since pads need a pressure to work properly.
 

·
Iconoclast
Joined
·
32,407 Posts
Even 21Gbps GDDR6X isn't that thermally dense. The reason it tends to run hot is because memory and VRM cooling on graphics cards is usually an afterthought.

Reducing the oft excessive gap between cooler and memory package helps a lot, and I often wind up using copper shims for this...preferably sized so normal thermal paste is sufficient, though I'll put a thin pad between the ICs and the shims (your #2 approach), if necessary.

If the memory ICs don't have underfill already, adding some can increase the thermal conductivity to the PCB (as well as protect the board and ICs), which can help thermals, especially if you cool the back of the board.

Using silicone gasket material as a binder for homemade thermal putty is a neat idea, though putty should only be necessary if properly sized shims and pads aren't available. Also, if you're going to be doing a lot of this, it's probably wiser just to purchase purpose made putty. Something like TG-PP10 is rated at 10Wm/K and costs ~50 cents a gram.
 
  • Rep+
Reactions: wuttman

·
Registered
Joined
·
878 Posts
Just got some TG-PP10. It has the consistency of modelling clay, very dense and relatively soft, but not oily / gooey like K5 Pro. Should work well for the application.

Planning to do the shim mod on a 1060 later tonight (ik, who tf cares about a 1060), hopefully it will increase the max stable mem OC. Going to use the first method that you suggested, and probably coat the caps around the edge of the memory with nail polish for good measure.
 

·
Banned
Joined
·
5,087 Posts
Could the thermal putty be used in place of thermal pads? The Phobya thermal pads I bought only have half the thermal conductivity of the thermal putty blameless recommended.
 

·
Registered
Joined
·
878 Posts
Could the thermal putty be used in place of thermal pads? The Phobya thermal pads I bought only have half the thermal conductivity of the thermal putty blameless recommended.
Yep. According to the datasheet, it's designed for 0.25mm - 8mm gaps.

Just got done doing the shim mod on a 1060. looks like it increased the max mem OC by 40MHz. Not that significant, but I assume the memory wasn't very hot to begin with. I would expect the improvement to be greater with GDDR6X. Forgot to take pics of the putty on top of the shims, but you get the idea. Chip > Kryonaut > 0.6mm shim > TG-PP10 > midplate

Also replaced the mosfet thermal pads with the TG putty.

2480403
2480404
 

·
Banned
Joined
·
5,087 Posts
I'd always wondered if cooling VRAM would increase the overclocking margins.

T.sharp is your GTX 1060 the one with the 6GiB of GDDR5x VRAM?

Do you apply thermal putty like thermal paste? Or do you spread it? Could you have just used the thermal putty by itself?

I believe the Xbox One X is using thermal putty instead of thermal pads.
 

·
Registered
Joined
·
878 Posts
@8051 Na, it's just a 3GB EVGA SC with GDDR5. Got to 6th place in Time Spy for the 1060 3GB though, 5th if you don't count the #1 score because it's bunk. 😄

The putty is more like soft playdoh and it comes in a jar, which makes it a bit of a pain to apply. I used a dab tool to get it out and just plopped a blob on top of the mosfets and shims, then compressed it with the midplate. I used more than I needed, but since it fills around the mosfets, I expect it to pull some heat out of the board too.

I definitely could have used the putty by its self though. Just didn't think it would have made much difference compared to the 6W/mK pads I had before. If you're upgrading from stock pads, it seems like a decent option.
 

·
Registered
Joined
·
878 Posts
Why not VRAM chip -> Thermal Paste -> Shim -> Thermal Paste -> Cold plate ?
It's definitely an option if you use shims that fill >90% of the gap. Using the thermal putty allows for effective cooling across larger gaps, so it makes sense if it's >0.2mm. It holds the shims in place more firmly than paste, but If you have a thin layer of paste on either side of the shim, it shouldn't move. I probably wouldn't run the card vertically if using only paste though.
 

·
Banned
Joined
·
1,320 Posts
Such Ram modules are not designed to get actively cooled, at DDR system ram, the heat-spreader this is not required but no harm at adding one.
Semiconductors top plate this is very thick, only 15% of thermal load this can be released due top cover.
VRM this is another story, their back this is an active heatsink and this is attached over PCB due soldering, on top heatsink this is added at their top cover as assistance because they have miniature dimensions, but the truth is that thermal pad this will absorb larger amount of heat from the PCB than the VRM cover.

In simple English, just compare dimensions of VRM VS Ram module, Ram module this is huge and does not use high amount of energy so to operate, therefore it does not need active cooling.
 

·
Registered
Joined
·
878 Posts
Such Ram modules are not designed to get actively cooled
Says who? The GDDR6X datasheet specifies 95c max operating temp. Even with active cooling (memory connected to main cooler with pads), modern cards can easily hit 110c memory junction temp. An overclocked 3080 can dissipate ~5W per memory chip. The cooler you can keep it, the higher you can OC.
only 15% of thermal load this can be released due top cover
Where did you find that data?
just compare dimensions of VRM VS Ram module
The actual memory die is much smaller than the plastic package. For GDDR5, it's ~5.5x5.5mm. About the same size as a typical VRM mosfet.
 

·
Banned
Joined
·
1,320 Posts
Where did you find that data?
You seem to know everything other than the thermal resistance (conductivity) of the plastic case.

modern cards can easily hit 110c memory junction temp.
There is no modern cards, what it is out there this is cards made with electronic parts placed at high density.
If you are up to to improve design, go and buy an 2000$ worth thermal imager.
With out it you are blind.

Orientation of VGA (PCB) within computer this is such that all components are reversed.
Its Electronic component this transfer heat at the PCB because heat this travel upwards.
 

·
Registered
Joined
·
114 Posts
Discussion Starter · #17 ·
only 15% of thermal load this can be released due top cover
This claim missing important variables to start making any actual sense. Let's say thick substrate that pcb made of has really lower thermal resistance than ~0.2mm? of plastic. That would mean, I dunno, 5 degrees drop on the back vs 15 degrees on top of plastic package. If you manage to remove all heat of the back side, it will surely prefer to travel that direction, but what cooling solutions can you apply to it? Mostly it's flat and passive backplate, that would be able to dissipate all the energy only at 100C+, which means most heat will be forced to go the other way, because 60+15 is better than 100+5. And looking at other people experiments with backplate cooling, thermal pads beneath is a serious bottleneck, since you can put 100W worth active cooler on the back and only drop temperature by 10 degrees. Sorry, but you're simply lacking knowledge to make valid judgement on benefit of using copper shims.
 

·
Banned
Joined
·
1,320 Posts
Sorry, but you're simply lacking knowledge to make valid judgement on benefit of using copper shims.
They are out there and far away from your tiny room, older smarter bright minds named as memory chip product designers.
Do not bother to contact them, they will never answer your email but they will do that at my own.
 

·
Registered
Joined
·
30 Posts
I did this mod (thermal paste - 0.5mm shim - thermal paste) and it decreased 10-12c from my asrock 5700xt challenger's vram temp. Awesome. I did have to sand the vram heatsink's spacers to reduce the gap, though. Would probably have been better with 1mm shims.

Before this mod, I had already reduced 10c by adding a 3mm thermal pad between the card and the aluminum backplate (also required some sanding of the spacers to increase pressure).

 
1 - 20 of 20 Posts
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top