My CPU temp sensor is not designed to actually report CPU temp?

[moeburn]

So, I have a motherboard CPU temperature sensor, but I also saw that 4 individual core temperatures were detected under the name "AMD Athlon II X4 620" in OHM in Windows, and the readout was precise to 0.1*c, so I thought surely that's a more accurate temperature reading for my CPU.  And sure enough, when I displayed it on a line graph along with the motherboard CPU sensor, it was like looking at a more precise version of the motherboard sensor line... 7 degrees lower.  So I just assumed the individual core one was right, and the motherboard one was wrong, and set a -7 degree offset to the motherboard one, and they lined up perfectly.

 

I had been doing that for years, thinking my CPU was running at 30*c idle, 45*c max.  

 

Then I read this:

 

 

K10-Tctl is a non-physical temperature on an arbitrary scale measured in degrees. It does _not_ represent an actual physical temperature like die or case temperature. Instead, it specifies the processor temperature relative to the point at which the system must supply the maximum cooling for the processor's specified maximum case temperature and maximum thermal power dissipation. 

 

Boy do I feel stupid.  Especially because I'm having a really hard time understanding all of that.  Could someone please explain to me what that means?  Given that the exact same sensor also reports a "high" value of "70.0*c", does that mean that if I add up the moboCPU-sensor, plus the average of the 4-core precise sensor, it should equal 70*c (it equals about 74)?  And right now it is specifically reporting a + positve value, shouldn't it be reporting a negative value if it is relative to the maximum temperature?

[Mindovermaster Moderator]

Sensors are guidelines, not the 100% true temperature. Your CPU can go to 40C to 45C to 30C back to 40C in milliseconds.

[PUC_Snakeman]

On my Dell XPS 730x, I run BOINC distributed computing on my 8 logical CPUs (I have a hyper-threaded 4-core i7 965 XE) and all 3 of my GPUs (GTX 660 Ti, GTX 460, GTS 240).

I monitor the CPU temps using Core Temp, and I monitor the GPU temps using Precision X. I have a tray icon for each of the 4 CPU core temperatures, and each of the 3 GPU temperatures.

My System usually runs very hot, and I try to control the cooling the best I can. I have water cooling for the CPU, I've added an extra system fan, and I use a custom fan profile within Precision-X to make the GPU fan go maximum speed at a point before the GPU hits 70 *C, so that my GTX 660 Ti does not ever underclock when computing or gaming.

 

The routine temps for my system components are:

 

CPU full load: 80-88 *C

CPU idle immediately after full load: 55 *C

CPU idle for 10+ minutes: 30-45 *C

 

GPU full load: 68-84 *C

GPU idle immediately after ful load: 55-70 *C

GPU idle for 10+ minutes: 30-60 *C

 

I am used to seeing the CPU temps IMMEDIATELY jump from 85 *C to 55 *C when load is released.

They also change from 55 *C to 85 *C IMMEDIATELY when load is applied.

 

Regards,

Jacob

[moeburn]

Sensors are guidelines, not the 100% true temperature. Your CPU can go to 40C to 45C to 30C back to 40C in milliseconds.

 

No no no what I'm saying is that THIS particular sensor is apparently not a guideline for CPU temperature!  It's an arbitrary relative number, a number of degrees Celsius relative to some temperature point (IE NOT the actual temperature relative to 0*c like a thermometer) I think, IF I understand that quoted text properly.

[the evn show]

Removed

[moeburn]

Put all of that together and what have you got? Well I'm not really measuring a physical temperature, but the value I'm returning is related to temperature. I'm not really reporting a core temperature either - but the value is related to that. The scale that I report the result on is somewhat arbitrary. On one CPU the max temp is 100C but it reports 80, on another the max temp is 80C but reports 80. It's not really a linear scale - temperature goes up, amount of cooling goes up, but they're not necessarily 1:1 related and I might even change this ratio for different kinds of CPUs.

Maybe I shouldn't call this "Tcore" or "Tcpu" because that would be misleading. It's more of a control value for the cooling system. Maybe I should call that "Tctl" for "temperature control".

That's the basic thought process that lead to Tctl.

See here: (search for TCTL Max) Tctl max is always 70 for AMD chips. tctl is 'how close' to 'tctl max' are you on this particular processor. If Tctl = Tmax then the cpu thinks it's overheating no matter what processor you have or what the actual maximum/current temperature of the CPU is. Also notice the 'thermal profile': that's the log vs linear scale I used as an example which would let some cpus scale up cooling faster than others.

 

Wow, thanks for that very informative post!  I ask all this because I am not using the BIOS's fan control, it spins too fast when idle.  I'm using lm-sensors and pwmconfig in Linux, and I was wondering which temperature sensor to bind the fans to; the imprecise (1.0*c) mobo sensor, or the precise (0.1*c) AMD Tctl.  I can guarantee you that Tctl is not on a logarithmic scale, because when I line-graph it and the mobo-cpu-temp sensor, it creates the exact same line that the mobo sensor does, it has the same slopes when the CPU temp increases and decreases, only the AMD Tctl is always exactly 7 degrees cooler than the Mobo temp.  

 

So if the AMD Tctl is reporting temperature in a linear scale, but not relative to 0*c, in theory I could just add a +7*c offset to the Tctl sensor, and have a much more precise version of my mobo sensor (I have tried this in the graph, the two line graphs of Tctl and mobo-cpu-temp line up perfectly when I add a 7 degree offset to one of them), right?  I assume I would not want to add the minus 7 degree offset to the mobo-cpu-temp sensor.

Also, I've told pwmconfig to spin the CPU fans at minimum speed (45%, since I can't hear the difference between 0% and 45%) when they are below 38*c and to spin them at near-max (90%) speed when they are above 46*c, do these sound like good values?  I've actually been using the AMD Tctl to do this for 2 years now, which is apparently 7 degrees cooler than the actual CPU temp, so it's actually been told to spin at 45% below 45*c and 90% above 53*c, but since I've never had any problems with this CPU, I assume that was okay too?