Temperature readings innacurrate on all sensors

[euniqe]

Hey everyone,

Motherboard is an Asus P8Z77-V LX. CPU is a Core I7-3770 Ivy Bridge. Running Windows 8.1.

Anyway with that cleared up, onto the issue. Speedfan, Ai Suite II both read CPU temps of 50 degrees (i'm currently converting video - CPU at ~40%). HwInfo reads significantly higher than that, as shown in the picture:

[attachment=1264]

As you can see, the HwInfo temperature readings are wildly contrasting to those indicated by the other programs. Now considering I'm using multiple programs, we can cancel out the possibility of a malfunctioning sensor - unless there is the unlikely chance both speedfan an Ai Suite II are broken, and not HwInfo.

Furthermore, If the CPU reached 90 degrees I would have probably noticed throttling to a certain extent (and it would be physically impossible considering I have a locked processor with a CM 212 Evo heatsink attached and the cpu is running less than half its capacity - should mention I forgot to add thermal paste during the first year of this cpu's life, and temps never reached 75c).

Anyway I re-installed the program entirely and no dice. Is there anything I can do? Not the end of the world, I was just hoping to use it in the OSD with RivaTuner.

And if you're wondering - yes I can turbo up to 4.3ghz with this processor. Actually, it runs at 4.43 ghz as base clock is utterly stable at 103mhz

 

[Martin]

I'm not sure what the other tools are reporting, but it might in fact be the temperature inside the CPU socket, which is always significantly lower. And what HWiNFO reports is the internal core temperature. You might check with tools like Core Temp if it reports the same values as HWiNFO. If you see a higher difference there, you might check if you haven't adjusted the Tj,max value in HWiNFO, or press the "Reset Preferences" button in HWiNFO settings.
If throttling has occurred is easy to determine from HWiNFO sensors - check the "Core #n Thermal Throttling" values.

 

[euniqe]

I'm not sure what the other tools are reporting, but it might in fact be the temperature inside the CPU socket, which is always significantly lower. And what HWiNFO reports is the internal core temperature. You might check with tools like Core Temp if it reports the same values as HWiNFO. If you see a higher difference there, you might check if you haven't adjusted the Tj,max value in HWiNFO, or press the "Reset Preferences" button in HWiNFO settings.
If throttling has occurred is easy to determine from HWiNFO sensors - check the "Core #n Thermal Throttling" values.

Cheers mate, I forgot about CoreTemp and the positioning of sensors. I'll check those things out.

 

[Chike]

Could the CPU be at 90c when package power is 327W when it is at 73c with 47W package power?

 

[Martin]

Temperature might not be likely in direct relation to power. It depends on cooling capabilities, etc.
Did CoreTemp confirm the core temperature values shown by HWiNFO ?

 

[euniqe]

Everything seems to be in sync now :/


Funny - all I did was go to sleep and wake up. (me, not the computer)

Sorry to waste your time - seems like it was just having a moment for a while.

 

[Martin]

No problem Sometimes things get better after sleeping a while

 

[euniqe]

False alarm


Forgot that I was stressing the CPU when the problem occurred, and I just checked it on idle, so I ran a stress test for 5 seconds to see what happened.


Core temp has a similar problem. It reads under any significant load temps above 90 degrees c - I couldn't read hwInfo as it froze due to the cpu load, but logging shows the same result.

However speedfan and ai suite are on the money, so I'm now baffled.

 

[Martin]

If both HWiNFO and CoreTemp confirm the same high temperatures then I'm sure this is the correct internal (core) CPU temperature.
The other tools most probably report the external (socket) temperature.
I think you should check your cooling - heat sink attachment, thermal paste, etc.

 

[euniqe]

OK, you're logic is entirely comprehensible - but I highly doubt my cpu is at 90 degrees.

As I said I have replaced (added) the thermal compound, really not that long ago. The CPU is locked. I have amazing cooling in my case (mobo never reaches a 5 degree delta, neither do any of my dedicated hardware temp probes except those near the exhaust fans) and I have one of the world's most popular if not the most popular heatsink. Considering it has no moving parts, this cannot fail and you can clearly see the fan operating properly. I have tightened the mounting screws - they are physically bottomed out as per the 212 evo design.

Look, i'm not going to cancel it out but I'm 99% sure the cooling is entirely fine. Besides - ai suite used to show up to 75 degrees before I added thermal paste. If HwInfo/CoreTemp is accurate, I should have cooked the CPU a long time ago. I also fail to see how there can be a 40 degree difference between core and socket temps when they are separated by about a millimetre or two.

 

[Martin]

90 C temperature doesn't still mean it will cook. Thermal limit is usually higher, you can determine this by right-clicking on the core temperature value and then "Adjust Tj,max." The value you see there is the temperature where the CPU starts to throttle. This should be displayed by CoreTemp too.
If you don't believe that your CPU is so hot (in the core) you can try even more tools, there are many other which can read the core temperatures.
But I think all will report the same. I have checked again the screenshot and the high temperatures are confirmed by other values like CPU Package and IA Cores.
If you still don't believe, then you might RMA the CPU or ask Intel support. Maybe the thermal sensor in CPU is not working properly, or the IHS is not properly attached to the CPU die.

 

[euniqe]

Yes but 115 degrees c means it will cook. Read my posts again.

 

[Chike]

I am not the OP btw, the wattage just caught my attention.

Temperature might not be likely in direct relation to power. It depends on cooling capabilities, etc.

My cooler (CNPS10X Performa) does not fall far behind the CM 12 Evo and temperatures are pretty linear to wattage.
http://i62.tinypic.com/2gwhtuu.png
On the left (I am going by package temp) (63 - 20) / (83.9 - 9) = 0.574 c/w average
On the right (95 - 21) / (132.9 - 8.4) = 0.594 c/w average
If I try to estimate the temps at 132.9w by those at 83.9w I get 92.4 pretty close to 95.
If I try to estimate temps at 327W by those at 132W I get 209c  :huh:

I don't think the 327W is right, it may be the same issue I had when I adjusted the BCLK?

 

[Martin]

I think that your problem was probably due to disabling SVID (rather than changing BCLK).
You also need to consider inaccuracy - for DTS sensors see: http://www.hwinfo.com/forum/Thread-CPU-Core-Temperature-Measuring-Facts-Fictions
Intel has never released more exact numbers about accuracy of power measurement in the CPU, but they say it should be used for informational purposes and not consider it as absolute and accurate.

 

[Chike]

I have never purposely disabled SVID, it was on auto though I will try and force enable on it, and report back.

My calculations may not be accurate, I would agree it may have been 180c, or even 170c and not 209c, but not 90, not if I get 95c at 132W, the CM12 evo is not that better than the Performa.

 

[Martin]

Sure, that power value (327 W) is definitively too high. Might be a flaw in the CPU, maybe the temperature is flawed the same way.
Sometimes a BIOS update (including new CPU microcode update) might help.

 

[parsec]

I've compared HWiNFO processor temperatures with those from the Intel Extreme Tuning Utility, while running its stress test. The maximum core and CPU temperatures were identical. If Intel does not know how to read their own processor's core temperature readings, who does?

I was surprised to see how close the Intel processor TDP reading was to the HWiNFO CPU Package Power reading. I've never seen a TDP/Package Power reading above 150W.

The distance between a thermal diode in a processor core and one in a processor socket are virtually very far apart. The internal thermal diodes are part of the processor core. A thermal diode in a CPU socket is... where? Under the socket pins? Next to the socket pins? We only cool a processor from its top, the bottom is not designed to dissipate heat. Does the thermal diode in a processor socket even touch the processor? If it does, it's touching the bottom, non-metal surface. So we have thermal diodes in two very different environments.

Would I expect a big difference between an internal thermal diode and one in the CPU socket? Yes.

If we check the OP's VCore reading when the cores are 86C - 91C, we see 1.400V. Hardly a low VCore, most people don't run their processors at that VCore when over clocked. For a non-over clocked processor, that is a high, Auto processor voltage. It does not need to be that high. That is the cause of the high temperature, not a bad reading in a monitoring program.

A midrange CM Hyper 212 on an i7-3770 at a VCore of 1.400V hitting 90C core temperature while stress testing the CPU? I believe it, no problem.

 

[Martin]

Thanks for the additional results.
I'm too not sure where exactly a CPU socket diode is placed, I believe it depends on mainboard manufacturer and their design. Maybe some vendors disclose more information about the exact placement, but I don't remember seeing such documents yet.

 

[forteller]

A midrange CM Hyper 212 on an i7-3770 at a VCore of 1.400V hitting 90C core temperature while stress testing the CPU? I believe it, no problem.

Add that to Ivy Bridge problems with IHS and I'm also not surprised with such high temperatures. As I happen to have the same CPU (3770K), I can confirm that temperature spikes to high 80*C region as soon as heavy load is launched (Intel Burn Test).
Here are my temperatures two minutes after IBT launch
[attachment=1295]
I would advise OT to lower voltage, as it is unnecessary for such small overclock. Running mine @ 1,3V degraded maximum stable frequency from 4,8 to 4,5GHz in 1,5 years.

Also, i dunno if that's a bug or not, but those high Power Values in OT screenshot doesn't surprise me. After 45 minutes of logging I have seen some strange min/max values frequently (for example 30000+MHz maximum core frequency in my laptop, or 10+GB/s download rate on my network card in my PC).