How does one review the effectiveness of thermal paste?

[xtreemchaos]

ive only ever used LM on laptops then as a last resort, with a desktop theres much more options when wanting to get it cooler. as for reviewing Tim its a hard one, ive watched loads of vids on youtube but thay differ so much but i do like the idea of the hotplate method as mentioned on here it would give a level playing field in my view. interesting post. thanks.

 

[DeathtoGnomes]

I'll drop this here, hehe. @lexluthermiester can prolly summarize, if needed.

Copper anti seize grease as thermal paste?

I gave this experiment one more attempt to see if I could get any different results. I found out that one of my favourite oil brands Penrite has a product called Copper Eze. This stuff looks much more like the Kopr-Kote where it's a bronze paste in appearance, rather than the Devon Stop-Seize I...

www.techpowerup.com

 

[lexluthermiester]

Who is going to try diaper rash cream?

While that would be interesting, I'm not going there..

 

[stinger608]

You don't even need to use a CPU to test thermal paste. If you are just looking to get the delta. Get a hot plate and a cold plate and measure the heat transfer between. Swap out thermal pastes to see difference. That is how it is done in development before it even touches a cpu/motherboard. You have a plate that can be electronically dialed in to very exact wattage loads. With very accurate thermal probes.

Bingo!! True thermal paste reviews are not even tested and reviewed with processors.

Trying to find a good true thermal paste review is pretty tough and far and few between.

 

[Bill_Bright]

I agree and that is why I find it crazy that some only last a few months.

What? Please show us a link to a TIM that only lasts a few months. Nothing anecdotal, please. Real studies or a link to the manufacturers page that show their products need to be replaced that often.

Even liquid metal will last years and years IF applied correctly - which is the tricky part, and when applied to nickel plated surfaces

 

[DeathtoGnomes]

Who is going to try diaper rash cream?

While that would be interesting, I'm not going there..

We need someone to guinea pig that.

 

[londiste]

Best way to test that would be using a Dummy heater. So that you can set a wattage that will always be consistent from 1 run to the next. Thing with using a real cpu is when you run a bench mark it can be slightly different from 1 run to the next. With least a dummy heater you can run say 200watts and it will be same heat from 1 paste to the next.

I don't quite get why this has fallen out of favor for cooler reviews. Hot plate heated up to specific W levels would be a very good comparison.

There used to be some test rigs like that some time ago. Someone (Anandtech?) had a heating element inside a briefcase. Quick Google found example from Silentpcreview's Fan Test System 2010:

 

[Kapone33]

Thats why, at least for Contactonaut, a huge red paper inside the package states the "Do not use with Aluminium" along with some Xs.

I use liquid metal for about 8~9months now. Nickel palated CPU IHS and copper coldplate of the AIO block/pump package. Its expensive compound not only for its price on purchase but also because you need to do several applications when copper coldplates are at least the 1 side of the 2 contact surfaces. If both sides are copper I assume you need to replace it sooner, at least in half time.

As of performance... yes it is the best TIM you can get. I saw about 5~6C reduction on max temp (100% load) compared to ArcticSilver5.
Does it worth it? That depends on how you see things. For me yes it does, and I knew from the biginning that with copper needs re-application after short(er) period of time and not just once. I dont mind the cost, its too small among other expenses.

I saw der8ouer talking about it saying that needs replacement after 2 months for at least 2~3 times before you leave it for long. I find that for the first replacement it needs to be done on the month, not 2. For the second its ok to leave it 2~3months.
I first replace it after ~50 days and it was a bit late. Gallium had saturated copper in such amount that whats left behind was almost like very thin layer of dried compound, like a thin leave.

View attachment 171570

View attachment 171571

View attachment 171572

It was a little hard to clean IHS but the AIO coldplate was a lot lot harder. I did some abrasion to remove it from copper surface from certain points/spots.

This is the final result.

View attachment 171573

Next application

View attachment 171574

After that I left it for 3 months and when I remove it again it was a lot easier for both as it wasnt dried this time. I'm on the 3rd month now (of 3rd application) and I will do it again next month. After that the copper saturation from gallium would be enough and slowed down for the TIM to last even longer.
You can see that AIO's coldplate is not as it was new, little scraped from "cleaning" process, but I did not see any cooling performance drop. I quess the high thermal conductivity of liquid metal makes up for it.

Another thing to consider other than cost, re-application and cleaning is the danger of shortcircuit components around the socket or the CPU itself. Needs extra care treatment, especially for beginners.

No thanks some Noctua NTH2 is all you need without the hassles. I use that on everything and it works great. I see 32 C as my idle for my 2920X @ 4.2 GHZ.

 

[Kapone33]

Amazon.com: Noctua NT-H2 3.5g, Pro-Grade Thermal Compound Paste incl. 3 Cleaning Wipes (3.5g) : Electronics

Amazon.com: Noctua NT-H2 3.5g, Pro-Grade Thermal Compound Paste incl. 3 Cleaning Wipes (3.5g) : Electronics

www.amazon.com

"recommended usage time on the CPU up to 5 years"

Yes it is only 1 year old but if it works like other TIM it is more for CYA purposes than anything else. I usually (nowadays) change my CPU every once in a while anyway. I do find though that you have to replace the TIM on a GPU much more often for good temps.

 

[Bill_Bright]

"recommended usage time on the CPU up to 5 years"

Yeah, and there's a "Best by" date on my bottle of Mediterranean Sea Salt too. Really? It's a rock!

MX-4 claims to have a "8 Year durability". Does that mean it suddenly stops working and needs to be replaced after 8 years and 1 day? No. It needs to be replaced whenever the cured bond is broken.

 

[EarthDog]

I don't quite get why this has fallen out of favor for cooler reviews. Hot plate heated up to specific W levels would be a very good comparison.

There used to be some test rigs like that some time ago. Someone (Anandtech?) had a heating element inside a briefcase. Quick Google found example from Silentpcreview's Fan Test System 2010:

This is a great method. The problem is a lack of understanding from most readers - the concept seems to go over their heads. People want to see how cooler A performs with CPU 1, not with a hotplate. That doesn't tell those users what they want to know.

 

[Bill_Bright]

People want to see how cooler A performs with CPU 1, not with a hotplate. That doesn't tell those users what they want to know.

Which is really naïve. The odds any review will be using the reader's specific CPU are slim to none. And a proper review is not about telling people what they "want" to know. It is about telling them what they "need" to know.

How is a hotplate any different from a benchmark program? Neither are actual "real-world" scenarios but do indeed tell folks the facts about how a certain product performs.

Same with testing power supplies in "hot boxes" - essentially temperature controlled ovens. Nobody runs their computers (and PSUs) in an oven. But testing a PSU in a 40°C (for example) hot box reveals information people "need" when deciding which PSU to buy.

If someone is going to venture into the realm of using TIM and going so far as to reading reviews, they need to do their homework, study and take the time to understand why that testing methodology was used, and use the results and conclusions to glean the information they "need" to buy the product that works best for them.

Also, read more than one review.

 

[Bill_Bright]

A hotplate is constant power; but transistors actually have less loss at higher temperature.

That has absolutely nothing to do with this discussion.

 

[Bill_Bright]

You realize this thread is about thermal interface materials, right? And how effective they are at transferring heat? If you do, then please think about your last two posts and how they are totally irrelevant when it comes to comparing TIMs.

This thread is NOT about "producing" heat. It is about getting rid of heat. Except maybe for the thermostat control circuit, a hotplate does not have any transistors. The block of metal (the heatsink) sitting on the temperature controlled hotplate, with just a thin layer of TIM in between, has no transistors in it either!

Therefore,

Your comment about power is irrelevant.​

Your comment about transistors having less loss at higher temperature is irrelevant.​

Your comment about diode drop is irrelevant.​

And your comment about some transistors taking advantage of anything is irrelevant.​

None of your comments provide any data whatsoever as to the efficiency of any TIM. And once again, this thread is about TIM.

 

[Zach_01]

I could agree with the argument:
"I want to use a hotplate to produce 100 or 150W of heat to simulate the X processor instead of putting the actual processor through a constant stress of 100% load"

Other than that, there isnt anything else.
TIM and cooling solution in general does not care at all (nor "knows") where the heat is coming from.
Heat is heat and all the same

 

[Bill_Bright]

I will let others judge if having a power source that does not vary is relevant or not.

Power source? Come on dude! It's a hot plate. The power source is immaterial and irrelevant. What matters is the hot plate having a consistent temperature. And that can easily be verified by the tester using a decent IR thermometer.

Heat is heat and all the same

Exactly!

 

[Bill_Bright]

OMG!

This thread is NOT about benchmarks. This thread is NOT about CPUs. This thread is not even about the temperature of the CPU.

This thread is about how efficiently a TIM can transfer heat from one piece of metal to another. That's no transistors, no diodes, no CPUs. You could even use the flame from a candle as your heat source. It would not matter as long as you can control the temperature of that heat source.

 

[Zach_01]

How about the argument:

A benchmark will produce the same heat from the CPU? (it actually depends on the temperature of the CPU)

Dont you ever run CPU burn/stress tests? Unlikely...

Every CPU if you load it with a benchmark/stress of constant load:

1. It will come to steady temperature within 1~2C after a certain amount of time (can vary)
2. If it is like a 150W power draw it will produce exactly the same amount of heat with a coldplate of 150W power draw.

Watts are used to mesure energy and heat also. Its the same. Heat is a form of energy.
Temperature is relative and has absolutely nothing to do with heat.

You have a CPU that produces 150W of heat
If you cooling the CPU with air cooler it will probably be around 80C
If you cooling the same CPU with the same 150W with liquid nitrogen at -100C (block temp) the CPU's temp will be like -90C

In both cases the CPU dissipates 150W

 

[Zach_01]

How many times must I apologize; now for it being unlikely that I run a stress test.

I ment that it is unlikely that you didnt ever run a stress test.

You are focusing on the wrong sentences. That was rhetorical question that was answered by me. I'm certain that you run tests. Please stop being defensive. If I had offended you I apologize.
Can we focus on the subject?
I wrote a bunch of other stuff there too, about heat.

The CPU produces less heat when hot (for the same task).

Can you elaborate this? How an object produces less heat when its temperature rises?

This:

Every CPU if you load it with a benchmark/stress of constant load:

1. It will come to steady temperature within 1~2C after a certain amount of time (can vary)
2. If it is like a 150W power draw it will produce exactly the same amount of heat with a coldplate of 150W power draw.

Watts are used to mesure energy and heat also. Its the same. Heat is a form of energy.
Temperature is relative and has absolutely nothing to do with heat.

You have a CPU that produces 150W of heat
If you cooling the CPU with air cooler it will probably be around 80C
If you cooling the same CPU with the same 150W with liquid nitrogen at -100C (block temp) the CPU's temp will be like -90C

In both cases the CPU dissipates 150W

EDIT:
typo

 

[Rei]

What the hell happened? Did OP's account got deleted or something? Did he just quit or banned?
And I was following this thread too...

 

[lexluthermiester]

What the hell happened? Did OP's account got deleted or something? Did he just quit or banned?

Very likely.

And I was following this thread too...

The thread is still going to work and the discussion can continue. I'm finding it very interesting.

We need someone to guinea pig that.

I'm going to pass.

Copper grease has the benefit of being made up of copper, which lends to the plausibility of it being a good thermal conductor. There is nothing about diaper rash cream that suggests it has any good thermal properties whatsoever.

 

[John Naylor]

1. You test TIM, the same way as anything else .... you select a single variable and keep everything else the same.

2. However there's a caveat... you can't keep **everything** else the same .... a cooler / TIM combo my do extremely well, taking top honors at unrealistic loadings ... and come in 2nd place or lower at more realistic loadings. You can see this in any one of TPUs cooler tests .... one cooler might come on top w/ P95 small FFTs and another might be best at AIDA 64.

3. Ideally, you should test things under the anticipated conditions of actual use. Since we are most concerned about what happens at the high end ... that means the highest anticipated situation that you actually might encounter running programs / games that you might actually use.

4. Whatever loading utility you use, it must be used consistently throughout the comparison for the same amount of time.

5. You also have to account for curing ... The makers of AS5 for example state that you need up to "200 hours of thermal cycling" to fully cure the compound. If we say that's normal usage .... at 4 hours a day, that's 50 days ... at 40 hours a week, that's still 5 weeks. last time I tested it and tracked the temps over time, we saw significant improvement in the 1st month, about 1/2 of that in the 2nd and in the 3rd it was only a tenth of a degree which is smaller than the sampling error i would presume.

6. I tried to do the "wayback machine" thing and find the old benchmarkreviews.com 80-way TIM test and the accompanying application methods article ... best treatise on the subject I have aver read. If ya can find it, give it a read.

 

[EarthDog]

1. You test TIM, the same way as anything else .... you select a single variable and keep everything else the same.

2. However there's a caveat... you can't keep **everything** else the same .... a cooler / TIM combo my do extremely well, taking top honors at unrealistic loadings ... and come in 2nd place or lower at more realistic loadings. You can see this in any one of TPUs cooler tests .... one cooler might come on top w/ P95 small FFTs and another might be best at AIDA 64.

3. Ideally, you should test things under the anticipated conditions of actual use. Since we are most concerned about what happens at the high end ... that means the highest anticipated situation that you actually might encounter running programs / games that you might actually use.

4. Whatever loading utility you use, it must be used consistently throughout the comparison for the same amount of time.

5. You also have to account for curing ... The makers of AS5 for example state that you need up to "200 hours of thermal cycling" to fully cure the compound. If we say that's normal usage .... at 4 hours a day, that's 50 days ... at 40 hours a week, that's still 5 weeks. last time I tested it and tracked the temps over time, we saw significant improvement in the 1st month, about 1/2 of that in the 2nd and in the 3rd it was only a tenth of a degree which is smaller than the sampling error i would presume.

6. I tried to do the "wayback machine" thing and find the old benchmarkreviews.com 80-way TIM test and the accompanying application methods article ... best treatise on the subject I have aver read. If ya can find it, give it a read.

1. Yep
2. A TIM's properties doesn't change in the meager temperature differences we see between tests. It is the cooler itself that shows a difference, not the paste.
3. See #2. You should test under worst case conditions/stress testing IMO as that will show where your peak is. Sure, you likely won't reach it with normal computing, but that also isn't the point of stress testing either. If we all bought for cooling gaming loads, we'd all have issues running anything else that is harder on CPU (most anything multithreaded).
4. Yep
5. AS5 improves like 1-3C. That said, few pastes out there need a curing time.
6. https://www.guru3d.com/articles-pages/guru3d-thermal-paste-roundup-2019,1.html

Best Thermal Paste for CPUs 2025: Over 50 Compounds Tested and Ranked

The right thermal paste will improve performance.

www.tomshardware.com