cheap but good thermal paste?

[eccerr0r]

Re: cheap but good thermal paste?

Well at least in my case, I think it made a difference for at least this one machine. Not sure how many people hammer their cpus for anything other than benchmarking, but as a Gentoo user my CPUs get thrashed quite frequently and for long periods of time...

[megaraider]

Re: cheap but good thermal paste?

Does thermal paste type really make that much of a difference? I'd think you would get more performance from replacing the fan on the heatsink or adding fans to the case. Unless we are talking about a laptop here.

Yes, see the chart in post #13. Upto 12c difference under load.

Even with a good thermal paste for a CPU running all cores at full throttle a 10ºC reduction won't be enough.
On a desktop the way to go was mentioned by clearchris :
Use a larger heatsink, change the heatsink fan, add fans to the case.
Making sure all the fans are rotating in correct direction is of most importance!

Well at least in my case, I think it made a difference for at least this one machine. Not sure how many people hammer their cpus for anything other than benchmarking, but as a Gentoo user my CPUs get thrashed quite frequently and for long periods of time...

IMHO can´t see why Gentoo makes a difference here, unless you're using AMD CPU.

[eccerr0r]

Re: cheap but good thermal paste?

It does ... if you're constantly updating Gentoo. No other distribution makes source update more configurable yet keep it simple enough that you can almost fire and forget. But source distributions requires a lot of CPU to finish these builds.

I figure most LFS users probably one and done it, and not sure about the popularity of other source code distributions... Most Linux are binary distributions and you don't need to compile each package.

And another observation: While trying to debug this same computer I swapped it for a larger case with faster blowing fans but it simply didn't help (note: the old case/fan system including the original HSF was working fine when it was new). Wasn't until after buying the new heatsink paste that the problem was finally solved - was kind of mad the old stuff that I did have seemed insufficient.

I think I will move everything back to the small stuffy case because it's quieter, now that the cooling issue is solved... The small case has a single, speed controlled 120mm fan; the larger case I was using as an experiment has 4 full speed 80mm fans...the inside case temperature is significantly cooler than the small case, but neither were sufficient to run four cores with the crap paste.

[clearchris]

Re: cheap but good thermal paste?

IMHO can´t see why Gentoo makes a difference here, unless you're using AMD CPU.

Heh, spotted the non-gentoo user.

If you run gentoo, you pretty much need to recompile a good part of your system at least once a quarter. I have given up on compiling the larger ones like firefox and take the binary distribution on those though. If you don't recompile at least once a quarter, you end up in dependency hell that can take a while to work through.

But as eccerr0r said, it's super configurable. You can set any build flag, compile in what you need, remove the rest and really be pretty sure it's all going to work. But it's not for noobs.

[eccerr0r]

Re: cheap but good thermal paste?

Aw come on, my Atom 1.6GHz only takes 25 hours to compile rust!

Incidentally I'm doing my quarterly update (actually, thirdly since it was 4 months ago) on my PVR box... Talk about keeping the machine warm for a while...

[stj]

Re: cheap but good thermal paste?

Atom's dont get warm, the shitty chipset does the heating for it!

[eccerr0r]

Re: cheap but good thermal paste?

they still can dissipate a watt or so, and that's sufficient to get warm. Alas my PVR is not my atom machine, it's a core2 and yes this gets warm.

[shovenose]

Re: cheap but good thermal paste?

https://www.amazon.com/Noctua-NT-H1-...p/B07MZ4JJWX/r

[megaraider]

Re: cheap but good thermal paste?

It does ... if you're constantly updating Gentoo. No other distribution makes source update more configurable yet keep it simple enough that you can almost fire and forget. But source distributions requires a lot of CPU to finish these builds.
I figure most LFS users probably one and done it, and not sure about the popularity of other source code distributions... Most Linux are binary distributions and you don't need to compile each package.

But as eccerr0r said, it's super configurable. You can set any build flag, compile in what you need, remove the rest and really be pretty sure it's all going to work. But it's not for noobs.

Yes my mistake.
Although I'm well aware it's a highly configurable Linux distribution, i misinterpreted and only took into account the use, forgetting how resources demanding it is to compile (much like any other similar).

And another observation: While trying to debug this same computer I swapped it for a larger case with faster blowing fans but it simply didn't help (note: the old case/fan system including the original HSF was working fine when it was new). Wasn't until after buying the new heatsink paste that the problem was finally solved - was kind of mad the old stuff that I did have seemed insufficient.

I think I will move everything back to the small stuffy case because it's quieter, now that the cooling issue is solved... The small case has a single, speed controlled 120mm fan; the larger case I was using as an experiment has 4 full speed 80mm fans...the inside case temperature is significantly cooler than the small case, but neither were sufficient to run four cores with the crap paste.

Even the best thermal paste becomes crap if you don't renew it!
The inside case temperature is a good indicator but, as we all know, not the only variable to take into account.

[eccerr0r]

Re: cheap but good thermal paste?

Well the problems *really* started when I took the HSF off to dust it with my air compressor I think (the machine had been overheating more often than expected hence the cleaning operation), and the overheating wasn't solved after several "fresh" ZnO paste applications (including mixing in some new silicone grease). But once I put in the CCP all is good once more. So I'm at least somewhat convinced at this point it was solely the CCP that solved my problems.

[momaka]

Re: cheap but good thermal paste?

There are two thing to consider when choosing thermal compound:
1) size of contact areas of the two surfaces
2) thermal dissipation (i.e. how many Watts of heat is expected to be transfered from one of the surfaces to the other)
3) heatsink size / thermal capacity / ability to remove heat

If the two surfaces have a fairly large contact area and the thermal dissipation is low to medium (i.e. an average 60-95W [true*] TDP CPU with a heatspreader and a heatsink adequate for the job... so not a stock HS in most cases), then any cheap thermal compound should get the job done just fine. In such case, the difference in temperatures between that and "premium" TC probably won't be more than 3-4C under load, if even that much.

Likewise, if the contact surface areas are small (i.e. exposed silicon die on a GPU), but the TDP is also relatively small (i.e. under 50-60W true TDP), then cheap thermal compound should also do the job fine here. And again, the difference between that and more expensive thermal paste probably won't be more than 3-5C in load temperatures.

* I say "true TDP" above, because Intel CPUs started abusing the TDP rating a long time ago, back since the P4 days, where TDP != maximum CPU power dissipation. Rather, they started using the TDP spec as an "average" power dissipation of the CPU under "average" load... whatever that's supposed to mean. Whereas with AMD, at least since socket 939 ans until socket AM2/AM3, TDP = maximum power the CPU can dissipate under full load.

Anyways, back to the discussed points above...
Now, if the contact surface areas are still large, but the TDP is very high (generally CPUs with TDP over 100W), then the difference in temperature under load will start to increase between cheap and better thermal compounds. Depending on the CPU TDP and surface area, what I've found from my own tests is that the difference can be as much as 5-6C.

And then there's the last case: if the contact surface areas are small (i.e. GPU with exposed silicon die), and the TDP is relatively high (over 60-80W), then it's best to go with a "better" formulated TC. Here, I've found that temperature differences can be as high as 6C or even more.

So, to sum it all up, use better/premium TC for CPUs/GPUs with exposed silicon die and high TDP and/or CPUs/GPUs that have heat spreaders but also relatively high TDP.

And while on the topic of heat spreader vs. no heat spreader, just keep this important note in mind: heat spreaders that aren't soldered to the die will generally tend to suffer from poor thermal performance, regardless of the thermal compound used. Case in point: an SFF PC I found recently with a Pentium G3260 CPU. Its CPU cooler was clogged with dust and probably caused the CPU to overheat a lot. This in turn likely caused the TC between the CPU die and the heatspreader to dry out. When I removed the cooler from the CPU, the TC between that and the heatspeader was already completely dry. So even after replacing the TC between the cooler and the CPU heatspreader, I still get dismal load temperatures, despite the Pentium G3260 being only a 53W [Intel] TDP -rated CPU (so probably true max TDP is around 65-70 Watts.) I tried both the stock and a taller Intel cooler, and I still got the CPU temperatures to touch 60C under load. So with that said, there's no point in using good TC if the TC between the heatspreader and GPU/CPU die is crap and/or dried up. Starting with the 3rd gen i3/i5/i7 CPUs, Intel really started using crap TCs... which is why many of their modern chips run so miserably hot. In contrast, most Xeon series and 2nd gen i5/i7 CPUs have soldered heatspreaders, as did the Pentium 4 and Pentium D (except for earlier P4 Northwood, I think)... which is why the latter can actually run really cool with large heatsinks, but the former won't get much of an improvement.

As for thermal compounds I use... I bought a 30g tube of some cheap no-name TC on eBay about 10 years ago or thereabouts (think I paid like $4 for it, including shipping) - likely the same stuff as that HY310/410/510 stuff, or whatever they're named. I've had really good results with this cheap TC on the following:
- Pentium 4 CPU (medium-to-high TDP, but still acceptable given the the contact surface areas and the fact that P4's use soldered heatspreaders)
- Pentium III and socket 462 Duron CPUs (small contact area, but also small TDP)
- socket 939 CPUs (poor thermal coupling to heatspreader due to it being not soldered, but TDP is still relatively low for the contact surface areas.)
- older GPUs with exposed silicon dies, but also under 65W TDP.

For hardware I care about more, I use MX-2, MX-4, and AS-5... with AS-5 being used mostly for high-power GPUs with exposed silicon dies.

And finally: if the contact surfaces areas are large, it's best to go with a more less viscous (i.e. "runny") TC, as that will allow it to spread into a thinner layer, which in itself will improve the thermal performance. On that note, this is where I *don't* like to use AS-5 (i.e. CPU with heatspreaders), as it's fairly thick and can't spread as thinly as more runny pastes... which in itself hinders some of its otherwise excellent thermal performance.

That's what I'm fearing, I just reapplied fresh ZnO paste on my cpu and it's still quite hot, wonder if changing pastes is worth the money or am I just tossing money down the sink...

Changing thermal compounds is always tossing money down the drain if the old TC hasn't dried up or been contaminated. With that said, I reuse thermal compounds between CPUs all the time, and I've never gotten more than 1C difference in temperatures between that and new TC, so long as I can keep the old TC from getting too contaminated with dust between transfers (not hard to do really.)

I've seen the HY410, HY510, and a whole bunch of other HY**0 formulations. Have you tried just the HY410 from multiple containers, have they been consistent?

I personally haven't... but I think it's just like Kingbo soldering flux - i.e. not the best stuff, and seems that there are different manufacturers that make it... but it's probably some generic flux formulation that's generally always good enough to do what it's advertised to do.

Does thermal paste type really make that much of a difference? I'd think you would get more performance from replacing the fan on the heatsink or adding fans to the case. Unless we are talking about a laptop here.

+1
When you start running into absurdly high temperatures, not even the best thermal compound can save you here. Every heatsink has a certain rated "thermal flow" capability. Go above that, and the temperatures will just rise. So yes, when the temperatures are too high, then a bigger heatsink/cooler would be the best solution.

[eccerr0r]

Re: cheap but good thermal paste?

Not sure how much of this is applicable here using a different/"bigger" heatsink/fan. Given a specific CPU and heatsink combination, and the fact it worked fine when new, this must be the baseline. After over a decade of use, the cooling solution became insufficient. As this is a 95W TDP CPU it has a lot of watts to dissipate.

Removing and cleaning the dust off of the fan and removing the original heatsink compound due to dust, fresh "Rawn" ZnO heatsink paste was applied - versus the original gray stuff that came with the HS/F - so two variables were changed, the thermal compound clearly was changed as well as the dust removal from the heatsink.

The ZnO plus heatsink cleaning simply could not match the original performance no matter how I varied reassembly which I would hope would not vary too much. So other than poor cleaning (used 90PSI air compressor which should be quite sufficient), it must have been the thermal compound.

White thermal compound is usually (cheap) ZnO and silicone grease. On the other hand, I still don't know what's in the gray heatsink pastes no matter what the manufacturer was. Would be interesting to see what's in this stuff - I suspect silicone grease is still the main carrier no matter what thermal compound we're talking about (other than the "liquid metal" ones) ... but the stuff mixed in seems to be some military guarded secret...

I highly doubt silver is used that much, which caused the initial hesitancy on buying new thermal compound as it would add significantly to the cost. Now that I find there are other grayish heatsink compound that's much cheaper than AS5 there must be something that can get better than ZnO without any silver.

[clearchris]

Re: cheap but good thermal paste?

Given a specific CPU and heatsink combination, and the fact it worked fine when new, this must be the baseline. After over a decade of use, the cooling solution became insufficient.

Suspect a dying fan. Most fans in my experience don't die outright, they slow down, or intermittently stop and restart when someone is watching.

I wouldn't have ever noticed this behavior, but when a similar fan goes out on my 3d printer, it results in a failed print, even if the fan stops for as long as 30 seconds... I also suspect that the fan performs differently with the cover off (no air pressure) and with the cover on. So you could have the observer effect, whereby actually observing the fan changes the behavior.

[eccerr0r]

Re: cheap but good thermal paste?

but... the system is working just like new once more now!

[redwire]

Re: cheap but good thermal paste?

Look at the ingredients, I find the base is also important for the stuff drying out. Most (electronics) thermal compounds are zinc oxide in silicone oil like:
MG Chemicals 860 Silicon Heat Transfer Compound
70% Zinc Oxide
3% Silica Powder
Base silicone oil?

Thermal Grizzly Kryonaut
65% Aluminum Powder
25% Zinc Oxide
10% Silicones


Non-silicone, like older Arctic Silver just ages like a paste clay, I'm not impressed with how long it lasts. 3 years is Noctua's spec.

Arctic Silver 5
?% Silver
5% Boron Nitride (aka white graphite, a ceramic powder, 3M comparison chart
30-40% Zinc Oxide
5% Aluminum Oxide
Base DiPhenylamine as "Synthetic polymer" Polyol Ester

Noctua NT-H1
60-80% Aluminum Powder
15-20% Zinc Oxide
Synthetic Oil Base

Coolermaster HTK-002-U1
30% Aluminum Oxide
20% Zinc Oxide
14% Silicon dioxide
3% Titanium Dioxide
1.2% Calcium carbonate
3% Magnesium Oxide
Base steric acid, palmitic acid

[eccerr0r]

Re: cheap but good thermal paste?

hmm... interesting. I'd expect aluminum powder to oxidize quickly and therefore become aluminum oxide.

How does MG Chemicals 860 work for most people anyway? I suspect my "white" heatsink compound is the same or at least similar to this...

And there's also the oddity that zinc oxide's thermal conductivity is higher than of aluminum oxide... so somehow it must be the aluminum that's conducting heat. The trick is to prevent the aluminum from oxidizing.

[redwire]

Re: cheap but good thermal paste?

MG860 is the old familiar silicone thermal grease, lowest performance because it's only zinc oxide I guess. MG Thermal grease flyer shows aluminum oxide is much better, get over twice the heat transfer with MG 8616.

Aluminum doesn't oxidize inside the grease (that keeps air out?) and it might be aluminum hydroxide particles already. It's something to do with the shape of the particles, spherical forms a longer path compared to anisotropic like boron nitride.

MG860 rated 0.66 W/m·K
MG8616 rated 1.8 W/m·K
Berquist TGR4000 4 W/m·K
Arctic Silver 5 rated 8.5 W/m·K (1-mil thick layer)
Thermal Grizzly rated 12.5 W/m·K
Noctua says the ratings are usually fake so they don't give any numbers for (Watts per meter-Kelvin).

standard ASTM test is with glass balls as a spacer.

I think for low cost, it's added aluminum to the zinc oxide. Copper is very good apparently but expensive and compatibility with metals on the heatsink/CPU is an issue.
Not sure why silicone vs polymer base is an issue, both seem to dry out.

[eccerr0r]

Re: cheap but good thermal paste?

pure ZnO should be around 60 W/(m°K). Pure Al2O3 is closer to 45ish and I think it's probably silly to use it. But yes it depends on the particle shapes, and the grease itself is probably the limiting factor - unfortunately it's needed else it wouldn't be spreadable.

Anyway it does look like it's probably the aluminum. I'm surprised graphite is not used. I guess I'm pretty convinced that gray > white at least for now, until I see some more consistently crappy samples.

[h0baz]

Re: cheap but good thermal paste?

artic sliver is also good

[bord]

Re: cheap but good thermal paste?

................
White thermal compound is usually (cheap) ZnO and silicone grease. On the other hand, I still don't know what's in the gray heatsink pastes no matter what the manufacturer was. Would be interesting to see what's in this stuff - I suspect silicone grease is still the main carrier no matter what thermal compound we're talking about (other than the "liquid metal" ones) ... but the stuff mixed in seems to be some military guarded secret...

I highly doubt silver is used that much, which caused the initial hesitancy on buying new thermal compound as it would add significantly to the cost. Now that I find there are other grayish heatsink compound that's much cheaper than AS5 there must be something that can get better than ZnO without any silver.

The grey stuff is carbon particles which is common in pastes. I don't bother about the contents, its not really important anymore.

AS5 is old and surpassed by many newer even cheap pastes out there like MX4, and its more difficult to spread and capacitative. But it has an advantage in that it does not 'pump out' as much due to its higher viscosity.

Pump out is a big problem especially for bare die cpus, gpus, laptops and to a lesser extent current cpus which have a higher heat density. Popular pastes like MX/NT-H1 are reported to have noticeably decreased cooling performance due to pump out, sometimes within a matter of weeks. Some of the old stuff like AS5, Ceramique, possibly the MX-2 seemed to be doing better in terms of durability.

The newer pastes like CM Maker Nano, NT-H2, Hydronaut, MX-5 are all more viscous to reduce the pump out effect. On the budget side, I'd look at the MX-2/MX-4, or Chinese made HY-810/880, GD900/900-1, they are made by Halnziye, Foshan HC.