Thermal paste matters! A review and data analysis

[Tek_Elf]

Some technical problems put me in the mood to do some general mantainence/cleaning on le ol' case, including dimm reoranging and mobo dusting. This required me to take of my monster of a heatsink and I had no paste; so this was an excuse to try this bulk buy thermal paste I had been eyeballing for months. I baby my hardware as far as temperatures go so later decided to go ahead and get a syringe of Arctic Silver 5 and use this as an excuse to do some data analysis for you guys. Following are the results of some testing. Enjoy.

Hardware:

1. Mobo: Gigabyte ga-990fxa-ud5
2. CPU: AMD FX-6300 overclocked to 4.5GHz at a max voltage of 1.464 (+.174 over stock, if I remember correctly)
3. CPU cooler: Be Quiet! Shadow Rock 2 -- View here: http://www.newegg.com/Product/Product.aspx?Item=9SIA68V21E0641
4. RAM: ugh... something something GSkillz 16GB at... 1600Mhz?
5. GPU: GTX-760 4GB edition
6. PSU: Thermaltake Toughpower simi modular 750W 80+ Gold
7. Windows 10 (albeit hacked enough to be mine)

The two pastes compaired:
Cooler Master Ice Fusion
http://www.newegg.com/Product/Product.aspx?Item=N82E16835103080
vs.
Arctic Silver 5
http://www.newegg.com/Product/Product.aspx?Item=N82E16835100008&cm_re=arctic_silver_5-_-35-100-008-_-Product

Method of paste application:

That is a VERY thin (thinner than the factory applications you see on stock coolers) layer of paste that I apply to both the processor and heatsink. Some people find this to be a waste of time, but I always get good results out of doing this.

Method of testing:
I took note of average temperatures and any spikes that may be significant while playing a few games normally. By "spikes" I am referring to temperature increases that happen at the same time as sudden lag or otherwise reduced visual performance. Otherwise, I paid special attention to how much the temperature fluctuates (that don't corelate to lag spikes) and I will rate these as follows:

1. none/no - temperature fluctuations that do not exceed 2C
2. mild - temperature fluctuations that average 5C
3. moderate - temperature fluctuations that average 7-10C
4. much - temperature fluctuations that exceed 12C

EDIT NOTICE: by temperature fluctuations, I mean BOTH POSITIVE AND NEGATIVE. So if an average was 50 with a fluctuation of "mild" that means it got down to 45 sometimes, not just getting up to 55. What I mean by 'average temperature" is the most common average between these highs and lows. Should have clarified this earlier.

Unless otherwise specified, every test was done with absolute max settings in the applications listed at 1080p resolution.

EDIT NOTICE: To clarify, the room temperature was roughly the same during the times of testing and increased temperature over time would have not been a major factor due to enough airflow through the house (i.e. I didn't "hotbox" test with my room closed off, getting warmer over time). Sadly I can't control for variables any closer than that at the time due to not having fancy pants equipment.

---

Temperatures:

Path of Exile
Ice Fusion: average of 51C, spikes as high as 72C; much fluctuation
AS5: 45C, average spikes were 55C; mild fluctuation
Difference: about a -7C average

Star Craft 2 (actually playing an arcade game called Special Forces Elite 5, which has tons of units)
Ice Fusion: 63C; mild fluctuation
AS5: 52C; no fluctuation
Difference: -11C average

Guild Wars 2 - A note about how I tested in GW2. As I did not want to spend the time playing it because I haven't' been in the mood, I did not actually go out and do what I normally do. I only went to the in-game PvP lobby, which is notorious for having terrible framerate and having relatively high, standing around/idle temperatures. I only did this to have more diversity in the testing.
Ice Fusion: 52C; mild fluctuation
AS5: 40C; absolutely no fluctuation
Difference: -12C average

Unigine Heaven benchmark (max settings except AA, which is only 2x as higher tortures my poor 760)
Ice Fusion: 57C; moderate fluctuation
AS5: 48C; mild fluctuation
Difference: -9C average

Unigine Valley (max with 8x AA)
Ice Fusion: 60C; moderate fluctuation
AS5: 48C; mild fluctuation
Difference: -12C average

The Elder Scrolls 5: Skyrim (with enough graphics and gameplay mods to convert any console fanboy)
Ice Fusion: 58C indoors, 66C outdoors; moderate fluctuation (indoor or outdoor; more likely outdoor however)
AS5: 42C; moderate fluctuation I just redid the paste tonight so this last bit here on Skyrim is not conclusive. I went to one of the most intensive areas in the game (modded Whiterun exterior) but that does not account for all the unique landscapes and the way they interact with mods.
Difference: presumed -16C average

EDIT NOTICE: I'm clarifying something here because people seem to be getting the impression from the way that I listed the numbers that it was common for the temperatures to average in the mid to high 60s while using the Ice Fusion paste. This is not the case. Star Craft 2 testing was the only test that had consistent temperatures in the 60s, 63c to be exact as listed above (yes, Unigine Valley technically did average "in the 60s" as 60 is, well, in the 60s). Other than that, the only time it got into the 60s or higher was during temperature spikes, such as during Path of Exile. So, to clarify, the overall average (i.e. average across all testing) for the Ice Fusion paste was 58-60C and the overall average for the Arctic Silver 5 is 48C (and that was before the "200 hour set in").

---

Analysis:
This is very interesting. In all the testing I've ever done, I've seen a trend that suggests that lower temperatures fluctuate more. This experiment falls painfully in line with that trend. There are three things that I believe are responsible for this:

1. Dynamic fan speeds that adjust according to core temperature (this one is kinda "duh")
2. The mumbo jumbo magic that is thermodynamic laws (in regards to heat moving away from the core)
3. Inaccurate readings at lower temperatures

Number 3 is only speculation based on two things: 1-idle temperatures are unrealistic, and 2-it would make sense that core sensors would be optimized at being accurate around the temperature range that the core is expected to operate, which might mean that it is less accurate outside that range. I've seen people comment about how this is probably the case and you can see this in idle temperatures. For example, I sometimes get 9C idle. Nine degrees??? NINE??? That's impossible. That's approaching half my bedroom's current temperature.

With that out of the way, two things to note: Arctic Silver 5 not only keeps the core cooler, it has significantly less temperature fluctuation. I believe this to be showing that there is a great ease of transition for heat from the core to the heatsink, while with the Ice Fusion paste more heat has to be built up for the transition to be smoother.

The Path of Exile and Unigine Valley testing is particularly interesting to me. PoE is notorious for how it was coded: badly. You can get it to run just fine with decent hardware, but the strain is unnecessarily high. In particular, I included running with a few friends that have some VERY laggy builds - the kind of stuff that has caused server crashes. The two main reasons behind such CPU strain in PoE is, I believe, the particle effects and the sound, which of course has to be uncompressed, rendered, etc, and in PoE sound activators can be spammed like mad with multi-target attacks. The particle effects has a similar scenario in Unigine Valley. I noticed the temperature go up and stay up during the rain scenes. Similar things can be said about Skyrim.

---

Conclusion:
Holy pah! So obviously yes the trend of better paste = cooler temperatures is true, nobody could ever dispute that of course. But what an unexpected turnout! A bit of perspective: I'm putting both pastes under the same pressure here, which is a lot. What gets me is just how well a paste as cheap as the Ice Fusion performed under the circumstances. The hottest it ever got was 75C and that only happened once; otherwise the average high was 72 in Path of Exile testing. That's not bad; really, that's actually pretty damn good! $20 for TWO HUNDRED GRAMS of paste that will perform well enough under these circumstances? If you're a PC builder, this is my recommendation - pick this up.

Of course, I'm still using AS5 on my rig. I baby my things because I can't replace them. But this has been a really interesting experience.

[bound4earth]

Even though a thermal paste is non electrically conductive, you should still avoid getting it on any of the PCB components. The only part that needs thermal paste is the GPU die and the CPU lid or CPU die. There is a reason that no good thermal paste has you apply the paste with say a business card or tool in a coating like this, it can introduce air, which can create a pocket of air and doesn't create any better coverage then putting a pea size or line on the die/lid and spreading the thermal compound with the heatsink.

Back on track... Does the 6300 run that hot? One sec Googling to see if anything changed since I last checked but FX chips Max recommended thermal should be around 62 degrees C. Unless something has drastically changed your temps are extremely high. My 8150 never exceeded low 50s on stock cooler. Although to be fair, the stock cooler is better then Intel's garbage, your aftermarket one should be much lower. Be back in a bit.

[Tek_Elf]

you should still avoid getting it on any of the PCB components

Emphasis on that layer of paste being VERY thin. It doesn't squeeze out the sides once I screw it down. When I took it off today to put on the AS5, there was none drooling out around the sides or on the PCB. I've never had the problem of doing it this way being messy. As for air pockets, yes and no. How you screw down the heatsink is more likely to determine if there's any air pockets. And while the die itself might be the hotspot, having the entire surface in contact is preferable. Like I said, I've always got great temperatures doing this.

Which brings me to that maximum of 62C. I would say that's AMD's ultra conservative limit because myself and others I've talked to don't take that seriously. I don't remember what temperatures I was getting on the stock fan, so I can't comment too much on it. But...

My 8150 never exceeded low 50s on stock cooler.

I can only assume you are running it stock in that situation. Keep in mind that my 6300 has a near-35% overclock on it and I've been running it like that for a year now. I have things like LLC and Coola n' Quiet on so if it doesn't need the OC it tunes itself down. But I game alot and 1 year of 35% overclocked gaming has done nothing to shake stability.

So no, I would disagree that 62c is any kind of max. I would say 60-70 is normal when you're putting it under that pressure. 80C? Start worrying. Over 85? Cut it out. I believe system shutoff is at 90C for the FX chips. Not sure though.

All that being said, if I were to keep this paste and heatsink but run stock settings, I'd never get higher than 45C. Ever. That much I know because I have ran stock with this cooler in Guild Wars 2 before.

[bound4earth]

Just checked the sheets and AMD still lists the FX-6300 chip's Maximum operating temperature at 62.5 °C. This doesn't mean you can never exceed that temp, but you should not exceed that temp for long periods of time like gaming as it can drastically reduce the life of the chip. FX chips are very different then Intel and previous AMD CPU lines. They have a much lower thermal threshold.

You might be reporting the socket temp. AMD chips have 2 temps, socket temp (should not exceed 70) and package temp (temp of cores should not exceed 62). They might use another word depending on the temp reporting software. Again this doesn't mean you cannot exceed these temps, but to say you can go above them because they don't shutdown til 90 C is just nonsense imo. You should not exceed 70 on the socket temp during normal usage or playing games. You can but you drastically increase the failure rate and the life of the CPU. Also yeah the temps tend to be wildly inaccurate at idle (low).

My current hardware:
1. Mobo: ASUS Sabertooth 990FX Ver 1.0
2. CPU: AMD FX-8370 overclocked to 5GHz max 1.445v stable - 4.8GHz at stock 1.38v stable (default voltage)
3. Noctua NH-D15
4. G.Skill 16GB @ 1600Mhz
5. GPU Gigabyte GTX 970 4GB Gaming OC Edition rev 1.0 (Overclocked 150 core 500 memory)
6. SeaSonic (can't remember the model) 750W 80+ Gold

I tested my CPU temps with the stock heatsink on my FX-8150 and I can report a drop of ~5-7 degrees if I remember correctly going from stock thermal paste to Artic Silver. I then tested out a new thermal paste called Diamond IC and was able to drop 7-8 over stock. I did not overclock on the stock heatsink and used a Hyper 212 Evo overclocking it 500-600 Mhz if I remember correctly. Temps reached 55 degree C during stress testing.

A year or more ago I read an 83XX and decided I wanted to crank up the OC and heard some great things about the binning on the 8370. I purchased a a Noctua NH-D15 and some Noctua Industrial 3000 RPM case fans (overkill I know now). Got a max overclock of just over 5GHz stable on air with temps ~61 degrees C during stress testing.

I run my case fans on a fan controller at idle and light to medium gameplay my fans spin around 800 RPM and it ramps up to about 1200-1600 if my CPU temp exceeds 50 degrees.

Again I have no experience with the FX-6300 and the 83XX has been binned to death and I kind of hit the motherload. Not many 8370 CPUs can hit 4.8GHz on stock voltage. Yeah I did notice your ~1GHz and maybe that is why your temps are so high. If you are comfortable with em that high then by all means do it up. I was just trying to throw you a bone. They do get up there a bit. When I talk about the FX-8150 stock heatsink it was not overclocked.
I shared my experiences just as a reference. I wouldn't want to overclock the FX line of CPUs with stock, I mean you could but you wouldn't get very far I imagine.

Edit: corrected some numbers.

[Tek_Elf]

I just find it so strange that they would list the max temp for that chipset so low. A hard cap/max of 62C seems unreasonable if you're going to be using it for anything stressful. The vast majority of the overclocking guides and forum discussions that I looked into when I first OCed my 6300 showed a trend of most people being between 60 and 70.

Ah, found a video of one guy who has some relevant numbers. He doesn't link any sources for the max temperatures he claims, which is unfortunate, but this is roughly in line with what I've seen in my own experience.

So he claims that AMD says that the 62C cap is for 8 cores and the 6 cores have a cap of 70. Again, I'm not sure how true that is, but it does seem that that is what AMD would claim based on my own observations. And that socket shutdown temp of 85-90 I'd say is absolutely accurate. I accidentally did that to my system once because I had a cable stuck in my fan when I shut my case and didn't know it.

Just a few clarifications:
I'm not saying that running at 70C is good. I find that too hot, it's why I got the AS5 after all. I'm just saying that 62 seems like an unreasonable cap for the 6300. It seems that there isn't that much room between minimal temperatures for that chipset and max. i.e. if you're overclocking you're unlikely to be below 60 (unless you have great cooling) but above 75 is starting to push it. That's only a 15C wiggle room.

My own 6300 has never been above 77. Ever (err, except that time with the cable stuck in the fan). I think 77 was the highest I ever got while experimenting with OCing the first time around.

Also keep in mind that the temperatures using the Ice Fusion aren't that high. Other than that spike that happened with PoE, the highest average is 60C (averaging across all listed that is). That's why I was so impressed by how well such a cheap paste performed. Of course I don't like risking it, so again that's why I got the AS5.

And now the average is down to 48 with the AS5. So i'd say if you're not going to push a chip as much as I am, then the Ice Fusion is great. But if you are, don't risk anything and spend that extra buck on something like the AS5.

[Calculatron]

To clarify, the general recommendation for AM3+ platform CPUs is to keep temperatures below 62C for the core temperature, and 72C for the socket temperature.

1.464v is a bit on the high side, so I am not surprised that the CPU is getting that hot with that cooler.

Anyway, good read! I appreciate when people take time to create and learn. (For science!)

[Tek_Elf]

Do you know anything I could do to get after those two readings independently? HWMonitor, MSI Afterburner, and CPUID all show the same temperature and don't even list a name. They just say "CPU Temperature" or "CPU1." For the time being, I'm just going to assume that the 60C average I had (48C average now with AS5) is socket temp, as that seems to be more likely.

Yeah, I would prefer not having it run with that voltage. But it makes sense that I had to bump it to that to get a whole 1GHz overclock. I originally had it running on a lower voltage buuut hour long prime95 runs and Server vs Server zerg battles in Guild Wars 2 starting throwing errors and the occasional BSOD. It's the only reason I didn't take it all the way to 5GHz. Now THAT would be amazing to see. So instead I use that voltage for 4.5GHz and have never got an error because of it.

[Calculatron]

Any chance you can post a screenshot of HWMonitor? I know it can have some weird names - I think it usually calls the socket temperature the "CPU Package" or some such. Gigabyte's own monitoring program (which I think is called EasyTune? It's downloadable from their website.) may show you as well.

Failing that, AMD Overdrive should do the trick. However, keep in mind that AMD Overdrive displays thermal threshold, not your actual temperature. So, as your CPU heats up, it will actually show you a shrinking number - this is because it is showing you how close it is getting to being "too hot".

AMD Overdrive's option aside, the best way to spot the difference between the Socket and Core temperature is how the numbers behave when at idle, and at load. The Core temperature, when idling, will probably be below-ambient, while the Socket temperature will be 10-20C above ambient. When the CPU is at load, the socket temperature will be 5-15C higher than the Core temperature, but the core temperature will more accurately tell you the temperature of your actual core. This discrepancy/weird idle readings happen because the Core temperatures are fed through a type of equation that mitigates some factors in reading the temperature - in other words, as it was explained to me, it is very accurately guessing the temperature of your CPU. .... And that is why AMD Overdrive switched to the Thermal Threshold model of temperature reading.

[Tek_Elf]

Wasn't sure how desktop screenshot just a window, so here's my right monitor.

I've never used Overdrive, but I might look into it.

And yeah, that sounds about like how I was guessing core temperature gauges work in my above comment a few posts ago.

[LordXenu]

Nice contribution to the community. As I really love your article and love how you shared your data. Anybody with an AMD soft spot would really be alarmed at the temps. My 8350 was water-cooled and never hit 55c with a few exceptions.But yea running 66c for long periods of time isn't healthy for your CPU, even if the TJmax is 72c.

Anyways yea AS5 is such a old thermal paste when it comes to enthusiast brands. Almost every modern aftermarket heatsink would come with AS5 quality paste with their heatsink, since Arctic Silver 5 came out circa 2001? I mean I still squirm a little when people still use it cuz of it's 200 hour set in time and laypeople are just running benchies like 1-7 days after the install. Pretty sure that cooler master paste is just toothpaste really honestly IMO.

I would spend that 15 bucks for the amd indigo xtreme or the coollaboratory ultra. Or if you're a traditionalist paster then MX-4 or IC Diamond. They have gone down from the insane price of 30-50 bucks back in the heyday to more realistic pricing of 15 bucks on amazon. I mean if you like drawing on your CPU or heatsink that Coollabortory Pro is for you!

Some control to your test for future reference, would be, room temp as the room could be higher the longer you test. Also measuring exhaust air with a 10-20 dollar Infrared temperature gun. Also playtime or game duration would be nice. Maybe even downclocking and running CPU on stock for the stock vs. OC temps comparison. It really changes people minds running 400-500mhz lower just for 5-10c lower temps sometimes.

As for enthusiast point of view I would replace that stock 50 cfm fan, with either a; corsait h80 fan, a silverstone FM-121 or Delta Ultra Kaze. If you want the legit look just jimmy-rig 2 fan, the Hyper 212 evo style and add an extra exhaust fan like the Silenx IXP7618, sure it's 38mm thick but 90 cfm and 18db goodiness or another high RPM fan you can stand with zip ties. Also you can epoxy/glue a few rows of ram heatsink to the opposite/top side of that cpu heatsink for even more cooling, I've done that with my Hyper 212 for a few degrees of cooling.

So yea AS5 compared to a top tier paste is about 2-5c on average contact to CPU to average users. Adding another fan would increase cooling by 2-3c and adding heatsink adds street cred for turbo charging your heatsink! haha. A fan pointed directly at the CPU socket and one on the other side of the motherboard will help too by another 2-3 degrees. That could be up to 6-16 degrees cooler before going for AIO liquid cooling or watercooling.

As for gaming goes POE isn't a graphics demanding game, I've ran it when I had overheating/shutdown problems with my graphics card back then with Final Fantasy for the PC. though the recent last 2 expansions really used up more GPU power now and better optimization, I used to be capped at around 75-120fps now I'm rocking max at 144 fps all the time except in parties. SC2 is CPU limited and actually slows down the longer you play or units you have sorta like memory leakage. Heaven & Valley are really good benchmarks no complaints there, skyrim also good.

Lastly I would mention the paste spreading, but yea I think enough people argue about it already and I can't really tell if that's razor thin spread of not. Honestly I'd rather want you to use either the 'x' method or the 1 grain of rice method better, because when the heatsink compresses, the CPU there's actually some metal to metal contact. That's how it conducts more heat that way, paste is only used to fill in air gaps, not used like butter on toast.

Skinslabs which is not defunct tackled it years back, here's a link: http://skinneelabs.com/prolimatech-pk-1-arctic-silver-matrix-and-cooler-master-icefusion/

[Tek_Elf]

Thanks for your feedback. I've realized that I should clarify what average temperatures there were across the testing because it seems people have the idea that the averages were in the mid to high 60s during the Ice Fusion testing. They weren't. They peaked in those temperatures (and in the case of SC2 a steady 63C was average) but other than that the overall average for the Ice Fusion testing was 58-60C. Still too hot for me, which is why I run the AS5 now, with overall averages now down to 48C (and that was before setting; I just applied paste and went right to testing. I believe I mentioned that somewhere...).

So when you take that into consideration, that's why I say the Ice Fusion did so surprisingly well. LOTS of stress for such a cheap paste to be put under. And same for the AMD CPU; doing great, all considering.

I also should have clarified how I controlled for room temp and play times, I agree. I believe I'll go back and add edit notices where I explain some of that info.

Yeah, there's a lot of this and that about paste spread methods. I personally prefer what I showed in the picture. I haven't done extensive testing myself because, well, that takes a lot of paste and time. But I enjoy doing this method and the results have always been fine for me. Yes, that is a "razor thin" application. It takes me a while to get it perfect, but like I said, I actually enjoy doing it this way.

[LordXenu]

[Tek_Elf]

I do not mean to sound bitchy when I say the following, but this is the fourth time I have addressed my prefered method of paste application in this thread alone. I obviously am quite familiar with computers and it's not like I don't know all the different application methods. I've stated more than once that this is what I am choosing to do.

And to clarify (again), I'm quite confident in the efficiency of what I'm doing. The amount of paste that @Logan used in the part where he used the credit card method is WAY (easily 2, maybe 3) times the amount of paste I have on my rig right now. It's not messy. I'm not worried about air bubbles. My temperatures have never been above what I expected while using this method. Hell, the whole point of this thread was that they were lower than expected. I expected the Ice Fusion paste to put me in the upper 60s/low 70s as an average. I expected to have to remove my overclock until I ordered better paste, which I was planning on doing anyway and simply decided to sooner so I could do this testing for the thread. Hell, maybe one day I'll take the time and personally test different methods. I expect little or no change, considering how damn good I am at getting those paper thin layers of paste on.

Again, not trying to be snappy but I see this happening every time I go on this website. Something will be brought up, addressed, and more people will keep bringing it up as if it hasn't been answered a zillion times already. It just clutters people's posts.

[Calculatron]

It takes something special to mess up thermal material application. Trapping air buttons can be an issue, usually if you take off the heatsink and then re-attach it without changing out the thermal material. The "pea method" is the most recommended at the moment because it has the least likely chance of having anything "bad" happen - although bad, in this case, just means something that isn't the most optimal performance. No one's computer was lost by a couple air bubbles between the heatsink and the CPU; the difference is usually just a couple degrees.

Linustechtips recently did a video that covered the topic of different styles of TIM application. It's not the all-be-all, but it gets the point across that you should have something between the Integrated Heat Spreader of the processor, and the Heatsink that you'll be using to actively cool it.

Tom's Hardware did a great write-up on Thermal Interface Materials, and even covered how Intel and AMD IHS are shaped different, as well as how the CPU die is orientated differently on the PCB. I think they still have the graphs of all their TIM tests up on their site.

[bryann816a]

honestly it may matter it may not. i believe ltt tried to prove that it didnt. i feel you should just do the rice method and live with it. all cpu coolers disapate heat. so as long as you've got the sucker plugged down good. theres nothing to worry about.

thats how i do it anyways. cbf to make everything perfect. you can't obsess about these types of things.

[Rango]

Wasn't sure how desktop screenshot just a window

alt + print screen

then just hit the crop button after pasting in paint :)

[Ctrl_Null]

if you have an older cpu/gpu def worth the 6 deg difference. i wouldnt go out of my way to buy something that would cost almost $20 after shipping for 6deg. most paste shipped with the coolers are in the lowest range.

but its you want to overclock the hell out of it. go for it!

while i have a super nice watercooling setup for my taste i stuck with arctic silver bc its highly rated and i got it for free bc of a sale.

[Zerophase]

Thermal Grizzly Kryonaut comes with a spatula and applicator tip. Though, that's designed to get full coverage on the inside.