CPU Temperature Monitors Only Go To -13C?

[NinjaNife]

I have just installed my first Phase Cooler and am using RealTemp and CoreTemp to monitor my temperatures. However, both seem to stop monitoring at -13C. Is this a limitation of my hardware, or the software itself? If the software, what program goes past -13C? I have a digital readout on my Phase unit and it says -74C, but I know that is the evap temp and not the actual CPU temp. I am using a MSI P55-GD80 motherboard. Any ideas?

 

[Athlon2K15]

most of the time people just get stand alone temp probes and say the hell with software

 

[slyfox2151]

the temp is stuck and wont be accurate at all at negative -13

 

[NinjaNife]

the temp is stuck and wont be accurate at all at negative -13

So are you saying it is a glitch, and is actually above -13C? Or that it can't read any lower than -13C.. And is that the software or the hardware?

 

[NinjaNife]

most of the time people just get stand alone temp probes and say the hell with software

How would that work? There is a temp probe on the evap, but how do you get one on the CPU itself (like inside the plate and all, or what)?

 

[NinjaNife]

I just felt around the clamshell, and it seems to be really hot.. Is this normal for Phase units?

 

[burebista]

So are you saying it is a glitch, and is actually above -13C? Or that it can't read any lower than -13C.. And is that the software or the hardware?

You have a simple formula for measuring core temps:

Absolute Core Temperature = TJMax - DTS

TJMax value is known (written in CPU MSR since Nehalem) and DTS is read straight from CPU on-die sensor.
So the only variable is DTS. If it stops at 113 you'll always read -13°C as core temperature (of course, assuming that TJMax is 100).

 

[NinjaNife]

You have a simple formula for measuring core temps:

Absolute Core Temperature = TJMax - DTS

TJMax value is known (written in CPU MSR since Nehalem) and DTS is read straight from CPU on-die sensor.
So the only variable is DTS. If it stops at 113 you'll always read -13°C as core temperature (of course, assuming that TJMax is 100).

So is there any way to change that (like changing the TJ Max in the RealTemp settings) so that RealTemp (or CoreTemp) will show the lower temperatures, or am I getting confused by all of this..

 

[burebista]

So is there any way to change that ...

Nope. Instead you can bug Intel for their crappy sensors.

 

[NinjaNife]

Nope. Instead you can bug Intel for their crappy sensors.

lol probably will do that. Thanks for your help.

By the way, do you happen to know what the correct TJ Max is for an Intel i7-875K? RealTemp has it set at 99 right now, but I think that is just the default, and not the actual...


*EDIT*
Also, does anyone have any ideas as to why my digital temp readout on the phase unit says -69C, but RealTemp says that the CPU is 23C (positive 23)? Is there really supposed to be that much of a difference, or do I possibly have the cooler mounted incorrectly (or is one of the temp readouts incorrect)? Thanks.

 

[burebista]

By the way, do you happen to know what the correct TJ Max is for an Intel i7-875K? RealTemp has it set at 99 right now, but I think that is just the default, and not the actual...

As I said before for Nehalem and latter CPU's TJMax value is written in MSR so no more guess. So if RT says 99 then 99 it is.

Is there really supposed to be that much of a difference, or do I possibly have the cooler mounted incorrectly (or is one of the temp readouts incorrect)?

First for your K CPU take latest RealTemp beta, it contains a fix for Core i5-655 and Core i7-875 with their adjustable multipliers.
Second, formula is simple RealTemp is only doing a extraction of DTS readings from that TJMax value. So if DTS is wrong so your final core temperature it is.

 

[mmoeller]

The Distance to TjMax that is read from the IA32_THERM_STATUS_MSR is only a 7-bit number. So the distance can have values only between 0 and 127. Your Intel i7 should have the TjMax value in an MSR as well. So assuming TjMax = 99°C is correct, the lowest reading you could get is -28°C. But there could be additional limitations in on the values showing up in these 7-bits. And of course the core temperature sensors are very unlikely to give any usefull values at these temperatures.

You could try to monitor the temperatures with the Open Hardware Monitor as well. It is open source, so if you know C# a bit, you can take a look at the code to see how the values are read and if there are any artificial limitations. The Open Hardware Monitor drops negative temperatures in certain cases from mainboard sensors, because usually they are bad readings.

 

[NinjaNife]

You could try to monitor the temperatures with the Open Hardware Monitor as well.

I downloaded that and tried it, but for some reason it doesn't show any CPU temperatures at all. It shows temps for my motherboard, graphics card, and even my hard drives, but nothing for the CPU (just the clocks and such, but no temps). Any idea why?

 

[mmoeller]

Can you save a report (File / Save Report...) and post or send it via PM? Then I can take a look and see why its not working (most likely because I was too stupid when writing the code ).

 

[mmoeller]

Thank you for the report (PM). I have taken a look why the Open Hardware Monitor does not display anything in this case.

It also reads a TjMax = 99°C from the MSR.

The 7-bit value of the Distance to TjMax is 0x70 in hex, or 112 in decimal (read from IA32_THERM_STATUS MSR). This would give also a temperature of 99 - 112 = -13°C

But the "Reading Valid" bit is not set in your case in the IA32_THERM_STATUS MSR. Thats why no value is displayed in the Open Hardware Monitor.

So I think this -13°C value has no meaning (or I read the documentation wrong).

At "normal" temperatures the Open Hardware Monitor should display a core temperature value on your CPU.

 

[NinjaNife]

Thank you for the report (PM). I have taken a look why the Open Hardware Monitor does not display anything in this case.

It also reads a TjMax = 99°C from the MSR.

The 7-bit value of the Distance to TjMax is 0x70 in hex, or 112 in decimal (read from IA32_THERM_STATUS MSR). This would give also a temperature of 99 - 112 = -13°C

But the "Reading Valid" bit is not set in your case in the IA32_THERM_STATUS MSR. Thats why no value is displayed in the Open Hardware Monitor.

So I think this -13°C value has no meaning (or I read the documentation wrong).

At "normal" temperatures the Open Hardware Monitor should display a core temperature value on your CPU.

So you mean that since the -13C value is not accurate, the hardware monitor recognizes this and just doesn't display it at all? So that no bad information is given?

 

[mmoeller]

there is an additional bit in the register where the -13°C is read from, which indicates if the temperature sensor reading is valid or not (that's at least what the Intel documentation says).

In your case this bit is not set (meaning the temperature reading is not valid). At normal core temperatures as seen on other systems the "reading valid" bit is usually set (meaning the temperature reading is valid).

The Open Hardware Monitor prefers not to show any temperature which is said to be invalid by the hardware. If you use your CPU at higher temperatures it should display a temperature once more in the Open Hardware Monitor.

 

[NinjaNife]

I just did a Prime95 test and found something strange.. It seems that 2 of my CPU cores run 20C lower than the other 2. Meaning that 2 of them are running 35-45C, and the other 2 are running 55-65C... Is my CPU possibly malfunctioning (yes those temps are positive; I OC'd the CPU so it would show more), or is this normal?

Core #0: 43C
Core #1: 54C
Core #2: 27C
Core #3: 61C

 

[unclewebb]

Intel's sensors are designed to trigger thermal throttling at 99C or thermal shutdown about 25C after that. They were never designed for accuracy, especially at idle or when using cooling that takes the sensors down below 0C.

For RealTemp, I decided to ignore the bit in the register that determines if a reading is valid or not. By ignoring this bit on the Core 2 CPUs, I found that I was able to get reasonably accurate data beyond 100C up to the thermal shutdown point between 125C and 130C. How accurate is anyone's guess but my logic was that an invalid temperature number might be better than no temperature number at all and then a user could decide if it made any sense. I think I'll add a warning so users know when the sensor is reporting data that it has determined to be not valid.

If you are running Prime95 Small FFTs on all threads of your CPU then in theory your core temperatures should be almost identical when the CPU has had some time to reach its maximum temperature. Most of the difference you are seeing is usually just sensor error. There is a lot of information about these sensors that Intel has never publicly released so all software is left guessing. It's an approximation at best. On some CPUs the approximation is very accurate while on many others, it is not.

 

[NinjaNife]

I normally use In-place Large FFTs, since they seem to stress the CPU most in a short amount of time. Not sure if that would make any difference.


*EDIT*
I did a Prime95 test and it said that a hardware failure was imminent. From what I can tell, the CPU overheated when I had an air cooler (Zalman CNPS9900 NT), and that is possibly why some of the temps are so different. I contacted Intel and they have agreed to give me a RMA for my CPU, so hopefully the next one will work better.

 

[unclewebb]

If Prime95 tells you that you had a hardware failure that means your CPU is not stable because it's not set up correctly. Usually some bios adjustments can fix that up. There's probably no need to replace it.

 

[NinjaNife]

If Prime95 tells you that you had a hardware failure that means your CPU is not stable because it's not set up correctly. Usually some bios adjustments can fix that up. There's probably no need to replace it.

I had thought of that, but it had overheated before, and I had an AMD 965BE that did the same thing, and a RMA fixed the error.. Not sure, but we will see

 

[DOM]

so how high of a oc have you tryed ?

and if it moves alot lik you said it might not be on there right

also do you know wut load it was tuned for ? did someone build it for you or you got it used ?

 

[NinjaNife]

also do you know wut load it was tuned for ? did someone build it for you or you got it used ?

I bought it used from "cdawall" on these forums. I believe it was tuned for a smaller load, but he told me he ran it well on an AMD Thuban with a good OC at -30C or something, so I decided it should be able to handle an i7-875K..

*EDIT*
Like I mentioned above, there is no way for me to tell (even guess) what temps I am actually getting, since the monitor only shows to -13C, so it might be -20 or even -50 for all I know. All I can really tell is the temp of the evap (from the digital readout which has a probe on the evap), and that says -74C right now (it went down to -69C at one point, but the CPU temps said positive 30C at that point).

 

[Arrakis9]

i think the best way your going to figure this all out is to invest a little on a temp sensor that will go down to the negatives you need and stick it in the cpu socket right next to the cpu that should tell you exactly how hot or cold its getting. my thought is the temp sensors just arnt reading correctly