According to the manufacture of AS5, there can be up to 200 hours
of break in time required for temps to stabilize. This claim has been tested the results are shared in this thread.Materials, software, and experiment
A delidded i7-3770k (over-clocked to 44x100 with fixed vcore) was treated with a fresh line of AS5 both under and above the IHS per the manufacture recommended procedure for application to Ivybridge i7 chips (see above link). The Intel stock HFS was then attached and the system was stressed using either mprime
(prime95 for Linux) or burnP6
immediately after the HSF was seated to establish as baseline (time = 0) steady-state temperature for each stresser.
Core temperatures, vcore, and CPU fan RPMs were measured using a simple shell script
harvesting the linux kernel's lm_sensors readings. Ambient temperature also also recorded using the average of two external thermometers.
Once the baseline temps were established, the experiment was conducted as follows: the stress tests were repeated every 2 hours
for several days. This means 42 min of stress (12 min for burnP6 and 30 min for mprime) followed by 78 min of idle all day everyday for 3 days. Then the interval was lengthened to once every 4 hours
(42 min of stress, 198 min of idle) for a few days. Finally, the interval was lengthened to once every 6 hours
(42 min of stress, 318 min of idle) for the rest of the experiment.
The resulting data were normalized to the average readings for the initial run in order to study the temperature decrease and rate of decrease.
Mprime version 28.5 x86_64 running 8 threads preforming a FFT of size 440k.
cpuburnP6 has no settings for multiple cores but it was run in 8 independent processes so all cores would be stressed equally. Conclusions
The temps basically stabilize after 8 hours -- no where near the 200 hours the manufacture claims -- in this particular test. The magnitude of the decrease was 3-4 C relative to the initial temps depending on the stress methods which is pretty good. How general is this result? The hallmark of a scientific experiment is a reproducible outcome. This is just an n of 1 but no plans are in the works to repeat the experiment. Perhaps others will do it and report back. Feel free to use this thread if you do.Extra detailsCalculating zero
Zero was calculated simply by stressing with the program (either mprime or burnP6) and waiting for the temps to stabilize. In the plot of core temp adjusted for ambient vs. time (seconds), the pink color indicates data points I excluded
since the CPU was coming up to temperature while the blue color denotes datapoints that were included. The higher temp/shorter line is the temperature profile for burnP6 and the lower temp/longer line is the profile for the mprime run. Basically, cpuburnP6 was run for 12 min and mprime for 30 min, excluding the first 400 sec from each one. The 2nd plot just shows the same data unnormalized for ambient temp and split out by individual core:Data
This plot shows the time since the HFS was seated (days) vs. delta temperature "dT" above ambient for each of the stress programs. Initially, they were run every 2 hours, but after 3 days or so, the interval was lengthened to every 4 hours, and finally to every 6 hours until approx 200 hours had elapsed. Both stress programs gave similar results with the average temp drop occurring sometime around 8 hours and stabilizing around 3-4 degrees less than the initially measured temps.
The first plot shows the average of all cores and the 2nd one shows the values for the individual cores.Data table
The data table generated by the script linked above is available
in csv format. A more elaborated table where I preformed some calculations needed for the plots (difference from ambient temp, which points are included or excluded, average for time=0 for each stresser, etc) is also available
in csv format should you wish to plot anything yourself.