TAT vs Coretemp logging

[Martinm210]

Just ran a little experiment logging my crystal fontz, TAT, and Coretemp all at the same time to feel out the differences in the results.

This was just a simple 90 minute idle run with a D-Tek Fuzion all stock

Here is the big picture, a bit of a mess...



And to get a better picture of a short period of time to see the fluctuations in TAT and CoreTemp:


Hmm I thought, it's been a long time since doing any statistics stuff, maybe I can make sense of this another way. So here that is:

First TAT


Then CoreTemp


In the end, I guess neither one is as consistent as I had hoped, no wonder people have a hard time picking a result by watching it.

Seems at least with this one test, CoreTemp is very slightly more stable as indicated by the smaller standard deviations.

What bothers me with both of them is the distribution to one side. The little spikes off to one edge of the mean should theoretically be the same on both sides. I can only guess there is some sort of wierd CPU internal interference causing this.

Not much you can do about it, but average it out I suppose...

I guess more important may be running the exact same test again and comparing water/core delta from one run to the next and seeing what the percentage of repeatability is...

Also noteworthy is how TAT gave a larger deviation between cores than CoreTemp. And they were all recorded at the same time, yet a good 1 degree difference there..

[Snyxxx]

Great charts Martin. The statistics really clean the variation up. I always try to watch and determine a mean in my head, but this is a nice way to present the info.

[Martinm210]

Thanks!

In the end they both vary up and down more than I would like to see, but coretemp seems to keep the core temperatures between the two cores very close. There is really no proof of which is more correct, but my gut is leaning towards coretemp.

Under load, TAT seems to indicate my two cores on the E6600 are several degrees apart, where coretemp notes they are only one degree apart.

I'll do a few repeatability tests, but I'm leaning towards coretemp for my tool of choice for reading and logging because of the core temperature difference issue with TAT. I might still use TAT 100% as the tool for loading the cores though.

[rockqc]

Nicely done Martin. I'm just in the process of building my first loop and your testing on various aspects of water cooling has been a wealth of information.

Very much appreciated mate.

[Xilikon]

Yeah, that's what I also noticed during my search for a accurate temperature monitoring software last year with a fellow [H] member. CoreTemp seems to be very accurate but I don't think it is accurate at low temps, only on higher temps.

I think the thermal diode is calibrated to be accurate between 50C and 100C so if we get outside this range, it may not be that accurate. This make sense because the first purpose of the thermal diode is to let the CPU throttle at high temps and shutdown if it reach a critical point.

[nikhsub1]

Martin, in the end it is going to make very little diff... I chose TAT and now must stick with it, for better or worse. I like how it exports the file nice and clean for me... in an hour there are about 900 entries per core.

[Martinm210]

Martin, in the end it is going to make very little diff... I chose TAT and now must stick with it, for better or worse. I like how it exports the file nice and clean for me... in an hour there are about 900 entries per core.

Yeah, the repeatability is pretty close. Going to run a couple more tests, but it's looking like logging for about a hour produces a water core delta that is repeatable to within about .3C from the first few tests.

CoreTemp = +-.30C
TAT = +-.27C

I guess that's pretty good, and much the same either one. The only big difference is the delta between cores...for whatever reason CoreTemp remains about 1C different and TAT is 3-4 even up to 5-6C at times. In the end it doesn't matter if they averaged.