CPU Temps - Is Anything "Normal?"

[whovous]

I definitely do not have a crucial need to calibrate sensors. There is a strong argument that I do not even have a crucial need to own a computer. I build them because I enjoy doing so. Calibration sounds interesting, so I'd like to give it a try. I can always give up if it proves beyond my skill set. I've only read the Tom's Hardware guide once, and that was on a cell phone screen on a moving subway, so I do not claim mastery of the subject matter.

I have a Zalman 9700, which is a tiny bit better than the 9500. The quoted passage suggests to me that I should add 8 to my ambient temp of roughly 23C to get 31C. That is 8C less than the stuck idle temp of 39C, which tells me to add an 8C negative calibration factor in RealTemp.

Where do I go from there? I guess the glib answer is that I return and read the article on a real screen with fewer distractions. So far I have only a vague grasp of the material.

I hope that Chike will continue the discussion with Psycho101. My sensor is stuck at 39 at idle, but it will go up when stressed. Does that matter?

[Acebmxer]

Here is how I have my set up in Real Temp.

Idle Calibration 0.0 0.0 -4.5 -4.5
Tjmax 100 100 109 109

For my current O.c at idle mine read

Temp 38 38 38 38
Dist TJmax 62 62 64 64

My first core will go down to 33c 2nd core stops at 35c the last to cores stop at 38 or 36 at default config.

The sensors are not supposed to show temps but are supposed to help the cpu from over heating. Meaning once Distance to TJmax = 0 or TJmax = 100 the cpu will start throttling down

[Psycho101]

It doesn't matter if your sensors unstick at a higher temperature. You need them to work at idle to be able to use the method detailed. All stuck sensors will unstick after a certain temp. Those that seem constantly stuck may well unstick at a temp that will either mean the CPU is a charrd mess, or a temp you haven't reached yet/while watching. The key to the method in the first article is that the sensors are both unstuck at the rooms ambient temp.

As for comprehension, I didn't mean the realtemp calibration article, as that's pretty straight forward. I meant the article I linked to detailing load calibration. There's alot more in depth information contained within, some of which isn't really necesary to know to be able to perform a load calibration.

Even calibrated, the temp readings aren't going to be 100% accurate. An improvement, maybe, but not a complete triumph. I quit really worrying about temperature myself ages ago, as it's pretty meaningless. As long as I'm a reasonable distance to TJMax, I'm not bothered, but that's just me. There's no way you can ever say with confidence "Core 1 is 53c" because in all likelihood, even calibrated, it's most probably not.

If asked about DTS sensors and how to provide accurate temperatures for idle and load on a CPU , and told to sum it up in one sentence, it would be : "You can't polish a turd".

[whovous]

Can I go off on a tangent here?

Realtemp, as indicated, never ventures below 39C.

The HW Monitor in ET6 shows two temps:
System - 30C
CPU - 25C

What is the connection between these THREE numbers? Or are there just two? Is there a way to use the ET numbers for calibration purposes? Did I miss something obvious in setting up RealTemp?

[Psycho101]

Realtemp = DTS sensors (one sensor per core)
CPU Temp = CPU Diode (one sensor)
System = Diode somewhere on the motherboard, maybe on or near the Super I/O chip.

The link I posted a couple of posts back discusses relationship between diodes temps and DTS. However, even the diode isn't a true measure of temp. Intel state that TcaseMax (if that's the reading you want) is measured at the geometric centre of the IHS, on its outside surface. This isn't where the diode is located. The diode can be +18/-10c off.

[b_rubenstein]

I built a system with a EP45-UD3L and an E5200. The idle temp after start up in HW Monitor is always 41c. The temp reading from the MB CPU sensor is 20c lower. The way I figured out what made sense was to touch the HS on the GPU (passively cooled and the same temp reading) and the very base of the CPU HS. The GPU feels warm and the CPU feels cold. I now ignore the CPU idle temp readings.

[nate42nd]

This is a VERY good read on this question. i learned a lot about temperatures from this.

I would highly recommend reading it.

Sticky: Core i7 and Core 2 Temperature Guide - Intel - Overclocking

[marsey99]

stuck sensors cant be fixxed or worked around but even "working" ones are only roughly right in most cases for idle temp.

a rough guide would be to add about 10c to your "cpu" temp as read from cpu socket to get an idea of what your cpu cores really are, you could add a couple more for poor cooling and couple less for great cooling but this will give you an idea of idle temps.

once you get to a certain level with air cooling you will not get any lowwer idle temp as this is dictated by case airflow and room/ambient temp, load temps will give you the best indication of how well your cooler does.

Prolimatech Megahalems and 12 Other Coolers for Intel Core i7 Overclocking - X-bit labs

is a good round up of current blowers and how they compare on a hot cpu.

[Psycho101]

Adding 10c to the CPU Diode temp is just as "accurate" as taking DTS sensors at face value, ie it isn't. Truth is there is no easy way to get good temp readings with out drilling a hole in the centre of the IHS and adding your own sensor, as [H]ard OCP have done.

There are certainly much better ways to use the CPU diode than just adding a random amount to it. There are severasl guides around for calibrating using the CPU Diode that will produce much more representative results. Sticky: Core i7 and Core 2 Temperature Guide - Intel - Overclocking This shows the perfect way to get a representative TJunction reading from the Diode's TCase.

Yes, granted, roughly speaking core temps will be higher by at least several degrees compared to the CPU diode, but as I say, just adding a number to it isn't good methodology.