Comparison of AS5 vs. MX-4 on Haswell i7-4790K

[graysky]

I compared Arctic Silver 5 to MX-4 between my new i7-4790K and my NH-14D. In short, I ran mprime (Linux64,Prime95,V28.5,build 2) using large FFTs with 8 threads and logged the temps throughout the run once per second using a shell script. Ambient temp which is very important to consider did not vary more than 2 F since the system was placed in my basement where it is very consistent. The digital thermometer showed 59F as the low and 61F as the high which is approx 1 C.

Run 1 was Arctic Silver 5 which had cured for approx 52 hours. I ran mprime as noted above for 4 hours.
Run 2 was with MX-4. I ran mprime as noted above for approx 1-1/2 hours.

Histograms in blue show the temperature distributions for AS5 and those in pink show the same for MX-4. The solid black line for each core is the average temp for each core. You can clearly see differences between the two of 2-4 degrees (allow for +/-1 C due to the ambient temp range). AS5 was the superior TIM in the test experiment.

Test system
Processor: i7-4790K @ 4.40 GHz (vcore 1.232 volts under load)
HSF: NH-14D with 120mm and 80mm fan running at max
Motherboard: MSI Z97 MPOWER MAX AC

EDIT: I have to totally invalidate my findings based on a configuration oversight: it seems that $HOME/prime.txt on linux dictates what size FFT mprime uses. I have found that using "large FFTs" as I did the the experiment allows for values of 128k-1024k which is a range that causes a given CPU various levels of stress. The trend is for smaller values to give more stress and as a result, higher heat. In short, I have no way to go back and see which FFT size prime was using when I compared these two. Here are my findings using the same TIM, but varying the FFT size. Each run is a average of 20 min of running.

I have since locked the FFT size to 400k and will repeat this experiment.

V24OptionsConverted=1
WGUID_version=2
StressTester=1
UsePrimenet=0
MinTortureFFT=400k
MaxTortureFFT=400k
TortureMem=0
TortureTime=3
OutputIterations=10000
ResultsFileIterations=999999999
DiskWriteTime=30
NetworkRetryTime=2
NetworkRetryTime2=70
DaysOfWork=5
DaysBetweenCheckins=1
NumBackupFiles=3
SilentVictory=0
AskedAboutMemory=1

[PrimeNet]
Debug=0

 

[doyll]

Were both TIM prints / seating the same? How much different were the room temperatures? Just remounting a cooler can easily give a +/-2c difference with same TIM. Without doing the same test several times to be sure the TIM seats are giving consistent temps, it's hard to be sure one TIM is 2c better than another.

 

[WaterKewl]

I have read lots and lots about TIM over the years. Proper application, best for air, best for water, easiest application, conductive, non-conductive. At this point, I am ready to say that as long as you aren't using something like a cotton ball, or peanut butter, as your TIM, it's going to be fine.

Cheers for the detailed test results though. Might I suggest running the test 4 more times with each TIM re-seating/re-applying between each test, tossing the best and worst result for each TIM, then averaging the other 3 to see if there is a clear winner? This SHOULD remove seating from the equation. ~2C seems to be within margin of error to me.

 

[EarthDog]

I have read lots and lots about TIM over the years. Proper application, best for air, best for water, easiest application, conductive, non-conductive. At this point, I am ready to say that as long as you aren't using something like a cotton ball, or peanut butter, as your TIM, it's going to be fine.

Cheers for the detailed test results though. Might I suggest running the test 4 more times with each TIM re-seating/re-applying between each test, tossing the best and worst result for each TIM, then averaging the other 3 to see if there is a clear winner? This SHOULD remove seating from the equation. ~2C seems to be within margin of error to me.

Exactly what we said at OCF yesterday. This appears to be single mount testing which went against other testing results (MX-4 is the same or better, not ~3C off) that had more empircal testing. I love it, but you need to remove the BIGGEST variable of all... the mount... as best one can.

 

[graysky]

@doyll - Ambient was explain in the first post.
@all - Agreed that mounting is a variable I didn't control for in this since only singlet data. I'm not going to do multiples on each TIM. Too time consuming.

EDIT: According to this detailed study, so long as moderate to great contact is achieved, variability of <1 degree C can be achieved.

 

[EarthDog]

Thanks for the info regardless graysky!!!

Any more overclocking guides coming out??!!!

 

[graysky]

Glad you remembered. Nothing planned but I am studying the claim that 200 h is required for AS5 break-in. Will post when complete. After 12 h of data it does appear to have dropped as is consistent with my prev. post about delidding my i7-3770k

 

[EarthDog]

It was my first guide I followed in the Intel world... and brought me into extreme overclocking (LN2), hwbot, and eventually reviews. YOU DID THIS TO ME!!! LOL!

 

[graysky]

It was my first guide I followed in the Intel world... and brought me into extreme overclocking (LN2), hwbot, and eventually reviews. YOU DID THIS TO ME!!! LOL!

I am glad you found it useful... knowing people are getting value from it makes the time I spent on it worthwhile.

 

[graysky]

Regarding the comments above about variability introduced as a function of mounting technique: according to this detailed study, so long as moderate to great contact is achieved, variability of <1 degree C can be achieved. So I think these data and the conclusion stands as valid.

 

[doyll]

Regarding the comments above about variability introduced as a function of mounting technique: according to this detailed study, so long as moderate to great contact is achieved, variability of <1 degree C can be achieved. So I think these data and the conclusion stands as valid.

Validity requires all data involved in testing and doing each TIM mount several times.
I see nothing about or showing multiple mountings or temperature repeatability in the link you supplied. .

 

[Aquinus]

Run 1 was Arctic Silver 5 which had cured for approx 52 hours. I ran mprime as noted above for 4 hours.

52 hours isn't enough time to let it cure if all it was doing was sitting there. For me to get normal temperatures I had to wait at least 72 hours and I had done several heating/cooling cycles using furmark/kombustor as well before it cured that quickly.

I think the biggest problem with compound reviews is that no one really lets the AS5 cure long enough.

 

[graysky]

I think the biggest problem with compound reviews is that no one really lets the AS5 cure long enough.

Since there is already a pretty significant difference between the two, and since the assumption is that AS5 only become more efficient over time up to a point, I don't see the point of this criticism. The conclusion is the same.

 

[Aquinus]

Since there is already a pretty significant difference between the two, and since the assumption is that AS5 only become more efficient over time up to a point, I don't see the point of this criticism. The conclusion is the same.

The point is that it's not an accurate reflection of the improvement that one would have over the other. Yes, AS5 might be better, but the question is how much better is it? The point is that temps could drop another 4 degrees by the time it's completely cured and that at every step of the way for 4 furmark cycles (up to 80*C for about 30 minutes) over the course of 72 hours and gradually noticed every time that to maintain 80*C I needed to dump more and more voltage into my GPU until it got to the point that furmark was holding stead at 80*C at 1Ghz @ 1.3v which is practically my max OC on my GPU when it was barely able to keep it at 80*C at stock when it was first applied, so the change over time is huge and that even between the two last furmark cycles, temps at stock dropped easily 4*C where before that it was closer to 8*C and between that one and the very first was closer to 15*C.

So the argument isn't that AS5 is bad, it's that your review doesn't do justice to AS5 once it's ready to go because with the amount of time you gave it simply wasn't enough to be ready and needed more time to cure which would result in better numbers for the AS5... at least that's my observation.

 

[graysky]

I am currently running an experiment to study the claims that 150-200 hours are required for AS5 cure time. I am only 40 h into it, taking data points every 8 h. I will post the results once I found that the system has reach steady state.

 

[Aquinus]

I am currently running an experiment to study the claims that 150-200 hours are required for AS5 cure time. I am only 40 h into it, taking data points every 8 h. I will post the results once I found that the system has reach steady state.

It also calls for several heating/cooling cycles in addition to the time. So I would run Furmark or P95 (depending on the device) for 30 minutes, then shut down the machine. Let it cool off then do it again a few times. That will cure it the fastest.

 

[doyll]

Accurate testing needs other pertinent data. and repeating the same test sequence several times to verify it's accuracy .. also data on air in room to include temp; barometric pressure and humidity .. as they can and do effect air's cooling ability and therefor results.

 

[Aquinus]

no barometric pressure and humidity data

That has negligible impact at or near sea level. You would have to be at really high elevations for this to make any difference and since he's testing them all in the same location pressure won't fluctuate so much that it won't give accurate relative results.

I do agree that several tests with a recorded ambient temperature would be preferable.

 

[vega22]

thanks for sharing your findings dude

 

[doyll]

That has negligible impact at or near sea level. You would have to be at really high elevations for this to make any difference and since he's testing them all in the same location pressure won't fluctuate so much that it won't give accurate relative results.

I do agree that several tests with a recorded ambient temperature would be preferable.

I live at sea level and my low & high barometric pressures for this year are 971.4mb / 1037.3mb. Humidity is always rather high with low & high for year being 24..3% / 87.9%. These are from my weather station in the garden.
How much difference that makes on a cooler/radiator cooling a CPU I do not know, but I suspect it is more than enought to change the CPU temp. If all other variables were eliminated we would probably see a significant difference .. more than 0.1c.

The recorded temperature needs to be at the cooler intake air, not the room in general. Believe me, they are not the same.
Testing needs to be done with several mountings with same TIM as well as several more than one test on each mounting.

 

[GorbazTheDragon]

I seem to recall several tests saying AS5 was absolute garbage for thermal conductivity...

 

[graysky]

Consider the results I presented invalid. I did not control for the FFT size (thought I did but later realized the defaults define a range). I edited the first post of this thread and will repeat.

 

[WaterKewl]

Consider the results I presented invalid. I did not control for the FFT size (thought I did but later realized the defaults define a range). I edited the first post of this thread and will repeat.

I'm grateful that you admitted a mistake, and are going to correct it!
Mighty big of you, most wouldn't have done that.

As for everything above I think we're beating a dead horse. There are a lot variables for testing TIM. Unless someone was going to provide the materials, facilities, and pay me to conduct the scientific tests that would require hundreds of hours to complete, I wouldn't get complete results. Woulda, coulda, shoulda... Including myself, I see a lot of people criticizing your tests, and not performing any tests themselves. I still stand by my earlier claim: No matter what TIM you use, with proper application, you're going to be fine. Results varying in a few C improvement only really matter to those competitively OCing.

 

[EarthDog]

If this is to be done better, he needs to do a few mounts for each TIM as well...

The FFT thing really doesn't make much of a difference so long as the test is conducted the same way/length of time each time as it spins through the FFT lengths just the same. The big thing here is the mounts.

 

[Aquinus]

The FFT thing really doesn't make much of a difference so long as the test is conducted the same way/length of time each time as it spins through the FFT lengths just the same. The big thing here is the mounts.

It does if it's not constant. The size of the FFT could impact thermal output because more or less circuits inside the CPU might be required to do each computation. It's best to keep that constant so it can be ruled out as a source of error. Anything done to test between coolers or TIMs really need to be done the same way for everything. It's differences in the process that enable abnormalities in numbers to appear.

So from a purely empirical standpoint, everything that can be made constant, should be. It will make the results more accurate across the board and as you suggested, multiple mounts also suggests multiple tests to validate those numbers. It may be tedious, but that's science™!