DIY Quick Guide: TIM replacement in 15 minutes. (excellent results)

[Torn Mind]

Water in isopropyl alcohol won't cause any harm if there are no ions in it to screw up the flow of electricity. Cleaning with it should be done by applying the liquid to the cleaning material first and then rubbing that on the PCB.

 

[Unoid]

I got some New TIM and my two 290's mining LTC could use it but I don't wanna stop their 4 week straight of mining lol.

 

[crashtech]

You never know, if you make the stop brief enough, the lower power consumption might eventually pay for the work stoppage. Lower temps equal lower current draw.

 

[Unoid]

I read the better tim really only prolong's how long it will take to reach max temp. not necessarily lower Load temps by hardly much.

 

[Schmide]

You never know, if you make the stop brief enough, the lower power consumption might eventually pay for the work stoppage. Lower temps equal lower current draw.

Huh? Lower temps is not equal to lower current draw other than a fan may spin slightly slower. Typical fan power consumption is less <4w so a few deg here and there is not going to add up t anything.

 

[crashtech]

Correlation between core temperatures and current draw at given clock speeds and voltages has been clearly established.

 

[Schmide]

Correlation between core temperatures and current draw at given clock speeds and voltages has been clearly established.

Duh Ohms law and Conservation of Energy begs to differ with you. Temperature has basically no effect current draw. (I=VR)

Thermal expansion does change power consumption, but not at measurable levels.

Seriously show me where the laws of physics are broken please.

 

[crashtech]

Can't say I am conversant in the Poole-Frenkel effect, but you can check out IDC's thread here:

http://forums.anandtech.com/showthread.php?t=2200205

 

[BonzaiDuck]

First thing I noticed, there wasn't any mention of the other components that might be cooled with a VGA-card's proprietary heatpipe assembly. If I recall both from experience and reading, some of the proprietary coolers cover those surfaces. But now that I think of it, the heatpipe cooler itself only had a base that fit the GPU die, so probably a minor point.

Second, I think the OP (and generally the rest of you folks who've made this sort of VGA-card TIM replacement) missed an opportunity here. It amazes me that people pick all sorts of TIM formulations that include aluminum oxide and silicon grease. Barring the conductive Ultra Pro or Coollaboratory liquid metal formulations or metal pads, the best material you can put between a heat source and heatsink base is diamond. Now let me qualify that.

In the CPU discussions about delidding, it was argued that micronized diamond might corrupt a film on the CPU die and lead to "copper migration." I hadn't thought about it at the time those delidding threads were active, but I had done just such a thing with a GeForce 8800 card for which I'd installed a TR heatpipe cooler. It never showed a problem over a period of some six years, and it's still running "tip-top." I also think folks from Innovative Cooling (producer of IC Diamond paste) weighed in on this, and their opinion was that this was a myth about micronized diamond on such surfaces.

IC Diamond or similar is worth as much as a 5C improvement in thermal performance, and only about 2C short of what the conductive metal formulations offer.

Now, that being said, with some TIMs it may be that the quantity applied is a significant factor, but in the case if IC diamond, you can waste money and goop it up to your heart's content, or simply spread it on the HS base with a razor blade or credit card, and on the GPU die itself with a (better recommended) credit card.

I never saw such fuss over these issues before!

I will say one thing: there are probably a few more worries about TIM replacement on a GPU die than there are for the CPU, so I commend the OP on the thought of it. In the case of the VGA card, you are taking apart an assembly not intended for that purpose by the mfgr. Never encountered one where it couldn't be easily done, but it's still something to think about.

 

[Schmide]

I posted in that thread and still believe something is not isolated in that test.

 

[crashtech]

I posted in that thread and still believe something is not isolated in that test.

And perhaps you have a better explanation for his results that you did not share in that thread, but whether it is Poole-Frenkel or a more banal explanation like voltage regulation problems, the net result is that more current was consumed with a hotter CPU. Whether or not it is appropriate to assume GPUs will behave similarly is another question, though my understanding is that they are similar enough to make that kind of association.

 

[Schmide]

Since I can't prove otherwise. I'll concede.

 

[KingFatty]

Yeah I think it's about the current leakage and how it's leaking more at higher temps? It's just a fundamental characteristic of semiconductor transistors. CPUs and GPUs are made of transistors, so they will leak and temperature rise makes it worse.

Here is, what's the technical jargon term... buttload of pictures of charts etc. showing the increase in leakage with temperature increase:
https://www.google.com/search?q=leakage+current+as+a+function+of+temperature&safe=off&rlz=1C1CHFX_enUS519US520&espv=210&es_sm=93&tbm=isch&tbo=u&source=univ&sa=X&ei=fpAOU--GN8Xv0gHvt4C4Bg&ved=0CD4QsAQ&biw=1536&bih=795

 

[crashtech]

Since I can't prove otherwise. I'll concede.

Well, to tell the truth, I did not even know this was a controversial thing even though it contradicts my basic understanding of electricity. I have accepted IDC's findings uncritically because I trusted his understanding of semiconductors to be pretty good. Too bad he is not around to clarify or elaborate, I certainly do not have the knowledge to defend his findings.

 

[BonzaiDuck]

Well, to tell the truth, I did not even know this was a controversial thing even though it contradicts my basic understanding of electricity. I have accepted IDC's findings uncritically because I trusted his understanding of semiconductors to be pretty good. Too bad he is not around to clarify or elaborate, I certainly do not have the knowledge to defend his findings.

I read that dissertation on the Poole-Frankel effect. I think it said that higher temperatures lead to greater leakage, requiring more voltage, leading to higher temperatures etc. But I don't think there was any real question about it -- it is an established principle.

But -- not to have a p***ing contest about it, maybe I misunderstand or someone is wrong. But I don't think the Poole-Frankel effect is "incorrect."

 

[KaRLiToS]

Getting off topic.

 

[tolis626]

First thing I noticed, there wasn't any mention of the other components that might be cooled with a VGA-card's proprietary heatpipe assembly. If I recall both from experience and reading, some of the proprietary coolers cover those surfaces. But now that I think of it, the heatpipe cooler itself only had a base that fit the GPU die, so probably a minor point.

Second, I think the OP (and generally the rest of you folks who've made this sort of VGA-card TIM replacement) missed an opportunity here. It amazes me that people pick all sorts of TIM formulations that include aluminum oxide and silicon grease. Barring the conductive Ultra Pro or Coollaboratory liquid metal formulations or metal pads, the best material you can put between a heat source and heatsink base is diamond. Now let me qualify that.

In the CPU discussions about delidding, it was argued that micronized diamond might corrupt a film on the CPU die and lead to "copper migration." I hadn't thought about it at the time those delidding threads were active, but I had done just such a thing with a GeForce 8800 card for which I'd installed a TR heatpipe cooler. It never showed a problem over a period of some six years, and it's still running "tip-top." I also think folks from Innovative Cooling (producer of IC Diamond paste) weighed in on this, and their opinion was that this was a myth about micronized diamond on such surfaces.

IC Diamond or similar is worth as much as a 5C improvement in thermal performance, and only about 2C short of what the conductive metal formulations offer.

Now, that being said, with some TIMs it may be that the quantity applied is a significant factor, but in the case if IC diamond, you can waste money and goop it up to your heart's content, or simply spread it on the HS base with a razor blade or credit card, and on the GPU die itself with a (better recommended) credit card.

I never saw such fuss over these issues before!

I will say one thing: there are probably a few more worries about TIM replacement on a GPU die than there are for the CPU, so I commend the OP on the thought of it. In the case of the VGA card, you are taking apart an assembly not intended for that purpose by the mfgr. Never encountered one where it couldn't be easily done, but it's still something to think about.

I think that some people must gather some money and collectively buy a Haswell system.Then delid and replace the TIM on the CPU and GPU dies with IC Diamond and then monitor the results for about a year.If nothing happens,it's a myth.I think that diamond particles will scratch the relatively soft surface of a die,so I wouldn't risk it.

Well, to tell the truth, I did not even know this was a controversial thing even though it contradicts my basic understanding of electricity. I have accepted IDC's findings uncritically because I trusted his understanding of semiconductors to be pretty good. Too bad he is not around to clarify or elaborate, I certainly do not have the knowledge to defend his findings.

Regarding IDC,where is he?I haven't seen him around here in a while...His research threads were epic!

 

[BonzaiDuck]

I think that some people must gather some money and collectively buy a Haswell system.Then delid and replace the TIM on the CPU and GPU dies with IC Diamond and then monitor the results for about a year.If nothing happens,it's a myth.I think that diamond particles will scratch the relatively soft surface of a die,so I wouldn't risk it.

Well, I'm not trying to encourage anyone for bare-die application, but I did it with two graphics cards, and they've been running fine for some six years. I might not have done it, if someone had given me this argument about the abrasive effect on a "film" covering the silicon. But even the aluminum oxide particulate in a lot of TIMs is an abrasive. Not as hard as diamond, but . . .

The thing about the diamond paste -- it doesn't pump out like thermal greases. Once you've put it under pressure of a heatsink assembly, the particles don't move or "go anywhere." But you're right: somebody should test it on a CPU die.

Also, I thought that the GPU die fabrication made for a "tougher" surface. Where did I read that or hear it? But they're meant to withstand higher temperatures, I think.

 

[geokilla]

So I got around to reapplying the TIM on my 7950 and I got some of the old, dry TIM stuck in the die and they're making contact. Will that be a problem?

Also noticed that my VRMs are air cooled... Stupid XFX.

 

[geokilla]

So I got around to reapplying the TIM on my 7950 and I got some of the old, dry TIM stuck in the die and they're making contact. Will that be a problem?

Also noticed that my VRMs are air cooled... Stupid XFX.

Bump.. Still didn't install it yet because worried.

 

[KingFatty]

Could you elaborate on what you mean by "they're making contact"?

It's confusing because the whole point of the TIM is to help make contact between the heatsink and the die. So if they're making contact, that's good and it's hard to understand what is bad about that?

 

[wand3r3r]

Do you have a picture? I was wondering what exactly you mean too.

 

[realibrad]

He is saying that he did not fully remove the old tim from the chip, and put new tim over the old one, and they are making contact.

This is not a huge deal depending on how much of an area you are talking about. Rule of thumb is to remove all of the old compound.

 

[geokilla]

Sorry I wasn't more clear. But if it's fine, then I'll just leave it.

He is saying that he did not fully remove the old tim from the chip, and put new tim over the old one, and they are making contact.

This is not a huge deal depending on how much of an area you are talking about. Rule of thumb is to remove all of the old compound.

You got half of it. Basically, I cleaned the die and removed the old crappy TIM to the best I could with 70% isotropy alcohol + QTIP, paper towel, and coffee filter. However, some of the old TIM got is now stuck in the edges and corners of the die in between the space where the brown things and the metal brackets are. The following photo taken off the Internet should help give an idea? After spending a good amount of time trying to get it out, I just gave up and put on new TIM, then reassembled everything back together. However I haven't used the 7950 yet just to make sure it won't short circuit cus of the TIM.

 

[wand3r3r]

Sorry I wasn't more clear. But if it's fine, then I'll just leave it.



You got half of it. Basically, I cleaned the die and removed the old crappy TIM to the best I could with 70% isotropy alcohol + QTIP, paper towel, and coffee filter. However, some of the old TIM got is now stuck in the edges and corners of the die in between the space where the brown things and the metal brackets are. The following photo taken off the Internet should help give an idea? After spending a good amount of time trying to get it out, I just gave up and put on new TIM, then reassembled everything back together. However I haven't used the 7950 yet just to make sure it won't short circuit cus of the TIM.

The confusion I had was since you referred to "die" but really you seem mean the whole GPU (in the picture including the green area and the metal parts connected to it) but not the whole red PCB. The GPU die itself (silicon AKA die) is just the small piece in the center (where the factory TIM was on top of).

I guess a couple concerns are if the TIM is thermally conductive. Does anyone know what those little parts are in the green area, it might matter if they could be shorted.

I would try clean off the green part but I'm not sure what those little pieces are.