Best TIM for GPUs

[Ajay]

There are plenty of tests of CPU TIMs out on the web, but I haven't seen any for GPUs.

I plan on putting a couple of GTX 460s in a new case soon. They will be running Folding@Home 24x7x365, so I figure I while I'm getting everything ready, I should replace the existing TIM (Which is probably crap) with something that will last longer and lower my temps by a few Deg. C.

TIA for your input.

 

[taq8ojh]

I don't know, I'd say the concept is absolutely identical, so if paste A is good for a CPU, it's just as good for a GPU.

 

[jaedaliu]

I don't care has posted a great comparison about thermal paste in the cpus and overclocking forum.

http://forums.anandtech.com/showthread.php?t=2285595&highlight=

 

[Ajay]

I don't know, I'd say the concept is absolutely identical, so if paste A is good for a CPU, it's just as good for a GPU.

I don't know if the compression forces are as great and hence, what the likely thickness of the TIM would be. If I don't get a better answer, I'll probably use liquid pro or try the liquid metal GPU pads.

 

[blastingcap]

http://forums.anandtech.com/showthread.php?t=2191561

Noctua NT-H1 apparently worked well in this epic GPU delidding thread. I think any CPU paste should work well on GPUs.

 

[Kenmitch]

When i put a water block on my 5850 I used ceramique II and performance was still rhe same years later.

My 7870 stays cool even pushing to extreme clocks. No need to even look whats in there currently. Guess I'll see what happens later as far as temps go.

 

[thilanliyan]

I started using NT-H1 after IDC's testing for my GPUs and CPUs...has worked well so far. I used to use AS5 for the GPUs and MX-2 for the CPU. Temps are a bit better now...like 1-3 celcius.

 

[BrightCandle]

As far as I can tell all the thermal pastes are within a couple of degrees of each other. There really isn't a whole lot of difference between them in terms of performance. There can be a reasonable difference in the life of the paste and in how easy it is to spread so its usually a good idea to get something that isn't totally weird.

I too use Ceramique 2, its nearly the top performer but more importantly it doesn't degrade and its easy to apply. Easy to apply means its consistent to apply. Fifferent applications can really differ in a lot of other pastes, so that is always a key consideration especially if you don't want to apply 2-5 times. Then you have to work out what the best performance is and then keep reapplying until you achieve it. Then with some pastes you'll end up redoing 12 months later when the paste goes off.

 

[24601]

The only thing to keep in mind is that if you are using electrically conducting thermal paste, like Arctic Silver 5, you need to not get any of it anywhere where it can cause a short.

 

[flexy]

I just did a job using MX-4.

Reason: Non conductive and said to be one of the top TIMs. Otherwise, If i had had some AS5 left I would've used that, but I was out of good TIM, that's why I got MX-4.

Edit: After I removed the OLD TIM from the GPU (GK-104), there was smeared stuff all over those tiny "dot" contacts surrounding the CPU, it took me the longest actually to wipe those things clean since I did not know whether the OLD TIM used was conductive or not. You definitely do NOT want AS5 or anything conductive smeared across those traces.

 

[flexy]

The only thing to keep in mind is that if you are using electrically conducting thermal paste, like Arctic Silver 5, you need to not get any of it anywhere where it can cause a short.

AS5 maybe if you do a neat, flat "smear" job and then just "hope" it wouldn't spread out. Still, risky in my opinion.

 

[thilanliyan]

AS5 maybe if you do a neat, flat "smear" job and then just "hope" it wouldn't spread out. Still, risky in my opinion.

I used AS5 exclusively for my GPUs for about 10 years with no problems on any of the cards I used it for. I did spread it thinly and evenly with a credit card. Just gotta be careful, but I agree it is an extra thing that can go wrong. I have switched to NT-H1 now.

 

[crashtech]

NT-H1 seems to be among the top 3 or so TIMs, but it is also very "idiot proof," which is why I use it.

 

[Ajay]

Thank you all very much for your input. I'm pretty sure I still have some NT-H1 left over from my last Noctua HSF. If I chose a conductive TIM, I'd just cover part of the GPU PCB with High temp tape (you know, the translucent orange tape).

 

[Ajay]

Different applications can really differ in a lot of other pastes, so that is always a key consideration especially if you don't want to apply 2-5 times. Then you have to work out what the best performance is and then keep reapplying until you achieve it. Then with some pastes you'll end up redoing 12 months later when the paste goes off.

A plus to running something 24x7 is that pump out isn't really a problem, since there is no cycling of temperatures so no problems with differential expansion slowly pushing the TIM outward.

 

[blastingcap]

I can vouch for PK-1 being a very good paste. MX-2, MX-4 also very good.

 

[Ajay]

I can vouch for PK-1 being a very good paste. MX-2, MX-4 also very good.

Thanks, I know PK-1 ranked well in Skinnee Labs test :http://skinneelabs.com/2011-thermal-paste-review-comparison/2/

I'll have to take a look and see how it did in IDK's tests (if it's there).

 

[tigersty1e]

the other 2 tips I would add is to apply a fresh application when the card is cold. overnight works best. that way everything is less dense and you can get a tighter mount.

the other thing is to add rubber o rings to the 4 screws. no matter how tight you screw it down, the mount pressure will weaken overtime.. usually the PCB is the first to shrink from the repeated heat cycles combined with strong pressure. the rubber will help keep the mount pressure strong.


Posted from Anandtech.com App for Android

 

[tigersty1e]

the other 2 tips I would add is to apply a fresh application when the card is cold. overnight works best. that way everything is less dense and you can get a tighter mount.

the other thing is to add rubber o rings to the 4 screws. no matter how tight you screw it down, the mount pressure will weaken overtime.. usually the PCB is the first to shrink from the repeated heat cycles combined with strong pressure. the rubber will help keep the mount pressure strong.


Posted from Anandtech.com App for Android

of course I mean more dense and not expanded.

of course you know what I meant already, but this post is "in b4 the grammar and context nazis"


Posted from Anandtech.com App for Android

 

[Ajay]

http://forums.anandtech.com/showthread.php?t=2191561

Noctua NT-H1 apparently worked well in this epic GPU delidding thread. I think any CPU paste should work well on GPUs.

Just read this, wow, NV's IHS really sucks. But I don't really want to risk killing either of my GTX 460s.

 

[WaTaGuMp]

I hear the grey colored stuff is pretty good.

 

[Ajay]

the other 2 tips I would add is to apply a fresh application when the card is cold. overnight works best. that way everything is less dense and you can get a tighter mount.

the other thing is to add rubber o rings to the 4 screws. no matter how tight you screw it down, the mount pressure will weaken overtime.. usually the PCB is the first to shrink from the repeated heat cycles combined with strong pressure. the rubber will help keep the mount pressure strong.


Posted from Anandtech.com App for Android

Thanks, though, as I pointed out above, pump out won't be a problem for this setup.

I hear the grey colored stuff is pretty good.

What a genius! And I'm looking over a bunch of data for no reason

 

[darckhart]

i use ic7/ic24 diamond for everything and it's been working out really nicely. non-conductive, no need to worry about Al or Cu surfaces, easy application, nearly no cure time. unfortunately, it does "stain" the blocks. or maybe i don't clean well enough.

 

[nightspydk]

Arctic silver the poison of choice here as well.

Taking a vga apart the most irritating problem is finding the screws to unscrew. That varies with model. Sometimes you need to undo the backplate again it varies so maybe experiment or google. I find often the springscrews are the right choice.

//edit