Best long lasting TIM in 2017? MX-4?

Status
Not open for further replies.
I used AS5 last time and was surprised researching for Ryzen how AS5 in some tests has really fallen behind. Since there's a lot of good stuff today, I'll probably be using MX4 from here on out just based on their 8 year longevity claim. Having found people individually claiming 6 years at least in person, I'm sold! That's one advantage of being on the market for such a long time. Being able to set it and forget it is a great feature. I got my 4g tube sitting right here ready to go for tomorrow (build day). It's probably difficult for some people with the brand loyalty flaw to get away from Noctua products, they are good but miss the mark here.

I will likely be swapping out my Cryorig cooler for a Scythe when AM4 gets some better options for mITX setups like mine, but sticking with MX4 when that time comes.


I'm not loyal to any brand. I give all the brands a chance and see how well they perform under my testing methods. I also usually purchase TIMs on both performance, longevity and costs. This time around I used the noctua because i was too lazy to dig the other tubes out of the drawers, lmao!! MX4 will work just fine.
 
Do you have a favorite of the most expensive pastes (that aren't electrically conductive and don't scratch a heatsink/IHS as ICD7 can)? I have until Tuesday until my AM4 heatsink adapter arrives so I still have time to change from the MX4. :) The MX4 was shipped 2-day so it'll be here tomorrow but I can always buy something else. I reconsidered the Noctura stuff when Kelijan mentioned it.

I have no preferences or favorites here... last time I used Arctic Silver 5 because in 2008 it was the hot new thing. Works fine but ready to try something else, especially something that lasts a long time without reapplying. I think MX4 and NT-H1 both look pretty solid. Unless it's attached to a block or AIO, I'm not convinced as long as you have a halfway decent paste that it matters. Considering how well MX2/4 does in tests across the web, the 8 years quote from Arctic has been the thing that pushes it over the edge into the winners circle for me.
I can't answer that because the systems that I have basically used the "set it and forget it" application of grease with have been primarily using Ceramique. Normally I save the highend stuff for systems that I overclock and upgrade cooling with so I don't have long term data. Here is a thought however: back in the day when thermal adhesive was terrible in thermal conductivity, I would use Arctic Alumina Adhesive (epoxy) mixed with a better thermal grease in equal parts. This improved thermal conductivity of already above average epoxy and slightly weakened the bond so that I wouldn't rip off BGA chips if I ever wanted to reuse the RAMsinks. This could be a good long term solution if you are seriously not going to be removing it for a decade.
 
I like Gelid GC Extreme. Consistently places very high, if not highest, in comparison tests.

Easy to apply. Lasts a long time.

I used to be a fan of Arctic Silver Ceramique. GC Extreme is better.

I concur. My GC Extreme application on my Xeon is now 2 years old. I've had no maintenance issues with the TIM. Clean up the dust every once in awhile and the temps return to initial application temp. Stuff is solid.
 
Built my mITX 1800X rig Tuesday night, all seems well with the MX4. I actually mounted it twice to see how it spread and how easy to clean off after using. Seems fine there, not overly difficult to remove with 91% isopropyl alcohol. Used the vertical line method for the first time, always did peasized blob in the middle before. Arctic Silver keeps pretty nice instructions for each platform and recommends the line for Ryzen.
 
I like Gelid GC Extreme. Consistently places very high, if not highest, in comparison tests.

Easy to apply. Lasts a long time.

I used to be a fan of Arctic Silver Ceramique. GC Extreme is better.

couldn't agree more
 
The whole "Intel and their crap TIM" thing is a pet peeve of mine...Everyone seems to know that going from crap to good TIM will only save you a couple degrees...yet somehow Intel's TIM is SO BAD that you can get a 30C jump by just changing the TIM. It makes no sense from a heat transfer point of view, or an engineering point of view, or even a marketing point of view. They're not saving millions on TIM. They're probably saving money by loosening the assembly clearances on the IHS to reduce the number of parts that fail testing.

You're mostly correct. A large contributor is the gap between the IHS and the die. This is caused by the adhesive used to hold the IHS keeping the IHS raised off die. Also the TIM used isn't the best in terms of transfer compared to alternatives, though in Intel's defense on this one, it is apparently extremely long lasting which is likely why that choice was made - they don't want thermal interface material expiring before the chip warranty.
 
You're mostly correct. A large contributor is the gap between the IHS and the die. This is caused by the adhesive used to hold the IHS keeping the IHS raised off die. Also the TIM used isn't the best in terms of transfer compared to alternatives, though in Intel's defense on this one, it is apparently extremely long lasting which is likely why that choice was made - they don't want thermal interface material expiring before the chip warranty.

I'm still curious how this happened. More specifically, did Intel intentionally make a change to the adhesive in order to increase the gap for some reason? Or was this basically an accident that they didn't fix because the chip is still operating within specs? That's something I doubt we'll get an answer on.
 
IC Diamond is garbage. Very difficult to apply correctly, best case temps are in line with MX4 anyway. I don't think I've found anything that's a better value than MX4 these days, and you can get it in large tubes too.
 
MX4 does not last very long grizzly or ICD works better for long lasting.

IC Diamond is garbage. Very difficult to apply correctly, best case temps are in line with MX4 anyway. I don't think I've found anything that's a better value than MX4 these days, and you can get it in large tubes too.
maybe you dont know what your doing?
 
I'm still curious how this happened. More specifically, did Intel intentionally make a change to the adhesive in order to increase the gap for some reason? Or was this basically an accident that they didn't fix because the chip is still operating within specs? That's something I doubt we'll get an answer on.

You've answered your own question. The chip performs within specified temps at the specified voltage and frequency. As far as Intel is concerned, it is WAD.

I am curious about Coffeelake. 50% more cores is ~50% more heat. They'll have to either A) reduce clocks (and take the PR hit for slower single-thread performance), B) improve the heat dissipation or C) accept lower yields and lower headroom and only ship what we'd consider "golden chips" or thereabout.

My guess is a mix of A & B. So long as it beats Ryzen, they can BS their way through any shortfalls in comparison to Kaby Lake.
 
You've answered your own question. The chip performs within specified temps at the specified voltage and frequency. As far as Intel is concerned, it is WAD.

I am curious about Coffeelake. 50% more cores is ~50% more heat. They'll have to either A) reduce clocks (and take the PR hit for slower single-thread performance), B) improve the heat dissipation or C) accept lower yields and lower headroom and only ship what we'd consider "golden chips" or thereabout.

My guess is a mix of A & B. So long as it beats Ryzen, they can BS their way through any shortfalls in comparison to Kaby Lake.
its called 50% more cores =50% more area so nothing changes. heat per area is the same so no change in temps
 
I've used many types over the years but I do tend to go back to MX-4 as it's a good all round paste. Not to mention the fact you can buy it in 20g tubes. I'm done buying stuff at 4g a time.

Just got my latest 20g tube a few days ago.
 
its called 50% more cores =50% more area so nothing changes. heat per area is the same so no change in temps

Except you're still potentially increasing the TDP of the die as a whole, and heat *does* travel horizontally through the die. The cores impact each other.
 
Except you're still potentially increasing the TDP of the die as a whole, and heat *does* travel horizontally through the die. The cores impact each other.

Heat does travel horizonally, but it's very minor effect. You can get a really good thermal model of the actual die by just literally using a 2D rectangle. There's just not enough "edge" relative to the size of the "face" to matter. The heat spreader does exactly what it's name implies and spreads out those point sources of heat quickly so that no one point of the die gets hot too quickly.
 
I've used many types over the years but I do tend to go back to MX-4 as it's a good all round paste. Not to mention the fact you can buy it in 20g tubes. I'm done buying stuff at 4g a time.

Just got my latest 20g tube a few days ago.

I actually bought a 20g recently and i've had to reapply paste a good 4 times already, worth it.
 
Heat does travel horizonally, but it's very minor effect. You can get a really good thermal model of the actual die by just literally using a 2D rectangle. There's just not enough "edge" relative to the size of the "face" to matter. The heat spreader does exactly what it's name implies and spreads out those point sources of heat quickly so that no one point of the die gets hot too quickly.

Your theoretical model doesn't quite match reality. I can max one core on my 7700K and watch the other cores rise at a rate of about 1c per minute up to about 8c over idle. Doing the same to 2-3 cores increases the temps in the remaining core(s) at an even greater rate and to a greater extent.
 
1C/Min is amazingly slow. And if you could watch that heat flow, much of it would be from the 'hot' core, into the IHS, where it's spreads out, and then back into the nearby 'cold' cores. The same effect occurs when you run more cores at once. That little steel plate is called a "heat spreader" for a reason.
 
Your theoretical model doesn't quite match reality. I can max one core on my 7700K and watch the other cores rise at a rate of about 1c per minute up to about 8c over idle. Doing the same to 2-3 cores increases the temps in the remaining core(s) at an even greater rate and to a greater extent.
you no longer have FIVR so those idle cores are not at .7v (or whatever) but are at 1.3v

Those temp increases would not happen like that if FIVR was still a thing. When one core increases voltage they all increase voltage even in sleep state because FIVR was removed.

Your reality is not seeing the whole picture of whats happening when you use a single core and your reality is misplacing why temps do what temps do.

So there is no issue with 50% more cores because it is no more dense than it was since it still is all on the 14nm node.
 
you no longer have FIVR so those idle cores are not at .7v (or whatever) but are at 1.3v

Those temp increases would not happen like that if FIVR was still a thing. When one core increases voltage they all increase voltage even in sleep state because FIVR was removed.

Test was conducted with c-states disabled and voltage set to manual, but thanks for playing.
 
Test was conducted with c-states disabled and voltage set to manual, but thanks for playing.

Well, my comments are based strictly on the thermodynamics of the situation, not the specifics of your settings. A single (or multiple) "hot spot(s)" on a silicon wafer will result in heat transfer TO the heat spreader, then back INTO the silicon. Silicon is a crappy conductor, and the wafer is super thin. The metal of the heat spreader will move it much more quickly to equalize the temps across the die.
 
Well, my comments are based strictly on the thermodynamics of the situation, not the specifics of your settings. A single (or multiple) "hot spot(s)" on a silicon wafer will result in heat transfer TO the heat spreader, then back INTO the silicon. Silicon is a crappy conductor, and the wafer is super thin. The metal of the heat spreader will move it much more quickly to equalize the temps across the die.

The exact mechanism is irrelevant. It's happening in real life and will have to be dealt with. On 4 core KL chips, the activity of the surrounding cores impacts temps by 8 to ~15 degrees.

If you think the heatspreader is the culprit, fine. That part isn't getting any bigger in Coffeelake. If it's the bottleneck now, it will be doubly so then.

It has to be accounted for.
 
The exact mechanism is irrelevant. It's happening in real life and will have to be dealt with. On 4 core KL chips, the activity of the surrounding cores impacts temps by 8 to ~15 degrees.

If you think the heatspreader is the culprit, fine. That part isn't getting any bigger in Coffeelake. If it's the bottleneck now, it will be doubly so then.

It has to be accounted for.

I'm not sure what you mean by "dealt with". On ANY chip, one core being at full load will cause the other surrounding cores to increase temp over time. GPU, CPU, APU, doesn't matter, it's just physics. The designers DO account for it, but there's not anything they can do to prevent it.
 
Test was conducted with c-states disabled and voltage set to manual, but thanks for playing.
either way it still is a non issue since total area and total power per area are the exact same.

50W over 1x vs 75W over 1.5x is still the same 50W over X. You are making an issue out of nothing.

The end result of temp will be the exact same over the whole area.

I dont intend to further reply because any further posting would be feeding a troll.
 
I'm not sure what you mean by "dealt with". On ANY chip, one core being at full load will cause the other surrounding cores to increase temp over time. GPU, CPU, APU, doesn't matter, it's just physics. The designers DO account for it, but there's not anything they can do to prevent it.

We seem to have lost the plot here. The issue is dropping 6 cores into a thermal system that struggles with 4. The single core test is only an example of how the additional cores *will* impact the other cores on the chip. Since we agree on that point, why is there such resistance to the conclusion drawn from it?

either way it still is a non issue since total area and total power per area are the exact same.

50W over 1x vs 75W over 1.5x is still the same 50W over X. You are making an issue out of nothing.

The end result of temp will be the exact same over the whole area.

I dont intend to further reply because any further posting would be feeding a troll.

Talking past one another is not trolling. You keep going on about the die. You are not wrong about it, it just doesn't matter. Only the wattage matters because the LGA1151 heatspreader is still the same size for KL and CL. Any heat from additional cores will have to dissipate through it. This setup struggles with KL's 4 cores, adding 2 additional cores will stress it further. That is the argument.
 
Last edited:
You are not wrong about it, it just doesn't matter. Only the wattage matters because the LGA1151 heatspreader is still the same size for KL and CL. Any heat from additional cores will have to dissipate through it. This setup struggles with KL's 4 cores, adding 2 additional cores will stress it further. That is the argument.


This right here shows you dont understand how this works.

The IHS size has no important impact on this and is not relevant to this discussion.

The only thing that matter is the area of the die and the amount of heat in that said area. If KL and CL are the exact same density the temps will be roughly the same with no major difference because the heat density is the exact same.

If there is a difference....if....it will be negligible. Should this shit just be soldered? hell yes but does it make a difference between KL and CL? No...not really.
 
We seem to have lost the plot here. The issue is dropping 6 cores into a thermal system that struggles with 4. The single core test is only an example of how the additional cores *will* impact the other cores on the chip. Since we agree on that point, why is there such resistance to the conclusion drawn from it?



Talking past one another is not trolling. You keep going on about the die. You are not wrong about it, it just doesn't matter. Only the wattage matters because the LGA1151 heatspreader is still the same size for KL and CL. Any heat from additional cores will have to dissipate through it. This setup struggles with KL's 4 cores, adding 2 additional cores will stress it further. That is the argument.

Well, a 6-core die will be physically larger than a 4-core die by all accounts and past history. So assuming roughly the same heat generated per core, you'll have more surface area to transfer heat to the IHS. So most likely you won't see any real change in temps on a 4 vs 6 core design.

Understand the heat spreader DOES even out the heat transfer over the die, but it's not a uniform temperature. And there's two rates to deal with. One, the rate INTO the heat spreader, which is determined by die temp the die size. And the rate OUT of the heat spreader, which is determined by the heat spreader size and the cooling system (so the temp on the top)

Unless of course you're currently saturating the heatsink, but that's only going to occur if you're using something REALLY underpowered. (like a low-profile stock cooler or similar)
 
Last edited:
I actually bought a 20g recently and i've had to reapply paste a good 4 times already, worth it.
I've used many types over the years but I do tend to go back to MX-4 as it's a good all round paste. Not to mention the fact you can buy it in 20g tubes. I'm done buying stuff at 4g a time.

Just got my latest 20g tube a few days ago.

I'm starting to think I should've gotten the 20G MX4 tube. Already used 1 of 4G in my current tube, and will likely be reapplying it again as I'm likely swapping out my 1800X for the 1700. Should be an 1800X for sale soon if anyone wants one at a discount (PM me if interested).
 
I've used many types over the years but I do tend to go back to MX-4 as it's a good all round paste. Not to mention the fact you can buy it in 20g tubes. I'm done buying stuff at 4g a time.

Just got my latest 20g tube a few days ago.

Haha yeah, I actually had no idea how much 20g is in size. So I ordered two. I was like "the hell, I'm never going to use all this!".
 
Used MX4 and AS5 in the past, had good experiences with no issues in terms of longevity since I upgrade about every 3-4 years or so. Trying TG Kryonaut as my paste TIM (GPU to GPU heatsink, IHS to CPU cooler heatsink) and TG Conductonaut as my liquid TIM (CPU die to IHS) this time around. Only time will tell how well they hold up, but I have a feeling both will fair just fine and outlast the usage-span of my builds anyway.
 
Coollaboratory Liquid Pro. It's the best performing compound out there right now that I know of (according to reviews, things run around 6C lower than MX-4 with air cooling), and since it's 100% metal it shouldn't dry up or anything. It's expensive, but worth it in my opinion. Or, for very slightly less performance, there's Liquid Ultra. It's a bit more expensive than Pro, but it's a paste so it's easier to work with.

But, if long life is your only concern I would go with something else (like MX-4) because Liquid Pro is a bit hard to work with and, of course, cheaper.

I wouldn't worry about it too much though. Anything that's not silicone is likely to last a very long time. My graphics card has Arctic Ceramique on it that's three years-old (from a syringe that's around 10 years old now and still usable) and the temperatures are the same as the day it settled. If I remember right, I had it on a CPU for about five years as well.

Point being, long lasting thermal compound isn't a new development by any stretch of the imagination and most manufacturers got on that bandwagon a long time ago.
 
Thermal Grizzly's Conductonaut and Kryonaut typically show SLIGHTLY better performance in comparison tests than Coollaboratory Liquid Ultra and Liquid Pro, respectively. Not a big enough difference to really matter though.
 
Last edited:
Thermal Grizzly shows Conductonaut and Kryonaut typically show SLIGHTLY better performance in comparison test than Coollaboratory Liquid Ultra and Pro, respectively. Not a big enough difference to really matter though.
I've never heard of those two. I must be out of the loop too. I'll be sure to check it out, thanks. :)

Edit: I see, 73k vs 80k thermal resistance. That sucks. I would have paid the extra $3 for the Grizzly. That might shave another .5 or 1C off! Unfortunately, I'm not kidding... lol
 
Last edited:
It's the new hotness due to the every so slight performance premium over CLLU and CLLP, so of course if you were ordering today it'd be a no brainer if the cost/availability is similar, but the difference is so small that it's not really worth regretting going for a different product in retrospect.
 
Been running MX2 for years. Is a couple degree C really going to make or break my OC when I'm well below thermal throttling? Not really.

If you have good cooling you'll hit the processors clock wall well before you hit a thermal wall regardless of which paste you use. I'm replacing my system or cooling after 3-4 years anyways, so 8 years of longevity isn't really important to me.
 
Well, a 6-core die will be physically larger than a 4-core die by all accounts and past history. So assuming roughly the same heat generated per core, you'll have more surface area to transfer heat to the IHS. So most likely you won't see any real change in temps on a 4 vs 6 core design.
Understand the heat spreader DOES even out the heat transfer over the die, but it's not a uniform temperature. And there's two rates to deal with. One, the rate INTO the heat spreader, which is determined by die temp the die size. And the rate OUT of the heat spreader, which is determined by the heat spreader size and the cooling system (so the temp on the top)
Unless of course you're currently saturating the heatsink, but that's only going to occur if you're using something REALLY underpowered. (like a low-profile stock cooler or similar)

How spectacularly wrong this ended up being. The 8700K hits 90C with a 280mm AIO. Compare with the 7700K.
DL5cadyV4AAA2Xh.jpg:large


You've answered your own question. The chip performs within specified temps at the specified voltage and frequency. As far as Intel is concerned, it is WAD.
I am curious about Coffeelake. 50% more cores is ~50% more heat. They'll have to either A) reduce clocks (and take the PR hit for slower single-thread performance), B) improve the heat dissipation or C) accept lower yields and lower headroom and only ship what we'd consider "golden chips" or thereabout.
My guess is a mix of A & B. So long as it beats Ryzen, they can BS their way through any shortfalls in comparison to Kaby Lake.
You nailed it. The only thing missing was, D) Pumped a ton of juice into it to maintain high clocks and let it rip. But certainly the results back up what you were saying in any case.

We seem to have lost the plot here. The issue is dropping 6 cores into a thermal system that struggles with 4. The single core test is only an example of how the additional cores *will* impact the other cores on the chip. Since we agree on that point, why is there such resistance to the conclusion drawn from it?
Talking past one another is not trolling. You keep going on about the die. You are not wrong about it, it just doesn't matter. Only the wattage matters because the LGA1151 heatspreader is still the same size for KL and CL. Any heat from additional cores will have to dissipate through it. This setup struggles with KL's 4 cores, adding 2 additional cores will stress it further. That is the argument.

I was rereading this thread and since Coffeelake has now launched, you were proven right. There's a few factors in play here, but ultimately the 8700K temps are absolutely through the roof.
 
meh.. granted it isn't over clocked but my wife is using basically my old pc and i5-2500K from 2011... I installed a Hyper212+ on it when I first built in in mid 2011 and it has been on there since with no removal ..etc just has had the case dusted out about ~twice a year.. runs fine.. it has been hot as heck here last few days.. so I was just double checking the temps on her computer (since obviously she isn't going to) and they are fine and ~near about what I remember them being in similar weather when I built this computer 6~ yrs ago... paste was Arctic Silver ceramique.

I had been planning on taking the HSF off and replying the paste and all.. and I still might but I'm in no rush after checking the temps out yesterday.
 
How spectacularly wrong this ended up being. The 8700K hits 90C with a 280mm AIO. Compare with the 7700K.

Most of the reviews I've been seeing haven't had temps that high. Guru3D's review was in the mid 70C range at load. LegitReviews was even lower on their temps. Hell, even the HardOCP review was WAY below 90C.

I seriously question that link you posted. Maybe their AIO isn't working quite right or they had a crap mount.

So, provide a bit more evidence before you start your slinging "spectacularly wrong" kid.
 
Most of the reviews I've been seeing haven't had temps that high. Guru3D's review was in the mid 70C range at load. LegitReviews was even lower on their temps. Hell, even the HardOCP review was WAY below 90C.

I seriously question that link you posted. Maybe their AIO isn't working quite right or they had a crap mount.

So, provide a bit more evidence before you start your slinging "spectacularly wrong" kid.

Middle-aged, grey haired kid checking in.:cool: If you looked hard at all the CFL reviews you would've seen this too. It depends what on load means as to the results. You can indeed hit 90C+, at stock with an 8700K. There were at least 2 or 3 review sources showing how much heat and a pig the 8700K is on power draw, from my memory alone. You just weren't looking hard, and didn't see the flaws. The top image from the two that were posted earlier is from LinusTechTips. You'll also find Ars Technica found the same, they were the source with the 280mm AIO hitting ~92C.

Anyway I'm not here to do that research, which I already did upon CFL launch all over again just to convince you. But that other guy you two were bashing was completely correct. In other words from what he was saying: there is no free lunch. Something had to give on CFL and it did, power/heat.
 
Middle-aged, grey haired kid checking in.:cool: If you looked hard at all the CFL reviews you would've seen this too. It depends what on load means as to the results. You can indeed hit 90C+, at stock with an 8700K. There were at least 2 or 3 review sources showing how much heat and a pig the 8700K is on power draw, from my memory alone. You just weren't looking hard, and didn't see the flaws. The top image from the two that were posted earlier is from LinusTechTips. You'll also find Ars Technica found the same, they were the source with the 280mm AIO hitting ~92C.

Anyway I'm not here to do that research, which I already did upon CFL launch all over again just to convince you. But that other guy you two were bashing was completely correct. In other words from what he was saying: there is no free lunch. Something had to give on CFL and it did, power/heat.

The reviews that I saw for 90C were all OC results, not stock. Everything else I've been reading shows it in line or bit warmer than the 7700K, but not by much. Linus was the only one with significant differences, which leads me to think there's another issue with his cooling.

And I'm not digging for flaws in any matter, since I don't own a CFL, nor am I likely to buy one any time soon.

I'm not sure why it's so important to do all this research to try to disprove a general point someone made months ago, but I guess you have lots of free time. The original discussion was about theoretical heat transfer for a fixed source across a fixed size and thickness plate (i.e., the IHS).
 
The reviews that I saw for 90C were all OC results, not stock. Everything else I've been reading shows it in line or bit warmer than the 7700K, but not by much. Linus was the only one with significant differences, which leads me to think there's another issue with his cooling.

And I'm not digging for flaws in any matter, since I don't own a CFL, nor am I likely to buy one any time soon.

I'm not sure why it's so important to do all this research to try to disprove a general point someone made months ago, but I guess you have lots of free time. The original discussion was about theoretical heat transfer for a fixed source across a fixed size and thickness plate (i.e., the IHS).

Yeah and it had real-world implications both parties here made future predictions on CFL, feel free to reread. All this research? I was rereading the thread as stated, and read the different lines of conversation contained within, and that guy (STR) got ganged up on... and he was right in the end. I'll always take a moment to congratulate someone in that scenario. Everyone dismisses you in the moment then forgets as soon as the results come in. I'm not doing any research, I found all of this out on CFL launch-day, just regurgitating the stuff I saw. It wasn't just LTT either, look around yourself but I already said that Ars Technica had the same at stock. Both are as credible if not more than the average site so dismiss at your own peril because it's reality and that means you're just punching upwards when trying to insist you're right.
 
Status
Not open for further replies.
Back
Top