"Core Temp" - A program for accurate K8 temperature monitoring

[mariosimas]

the new version of speedfan, 4.29, has also the core temp reading:

[The Coolest]

Yes it does.

[dickeywang]

Nice work.The temperature is about 1C higher what's shown in Speedfan. I consider it very accurate.



One problem though, why it doesn't show the TDP and TCaseMax for my Orleans 3000+?

[dickeywang]

Edit: Oops, sorry about the double post

[Praz]

One problem though, why it doesn't show the TDP and TCaseMax for my Orleans 3000+?

If I remember correctly 49C is not a valid value. If the processor reports this value I believe The Coolest decided to ignore it in this release.

[RPGWiZaRD]

I must say my TCaseMax and TDP value seems somewhat low for this chip... An energy efficient Opty 165... yea right.

EDIT: nvm I see it's an already known bug and meant to be fixed in next version.

[Lolento]

I just join to post this information regarding tcasemax and core temperatures...

tcasemax physically represents the maximum temperature allowed either at the top of the die or at the top of the IHS depending on how Intel or AMD did the thermal modeling.

core temperature since from the definition in the op is measured in the die, can only represent the junction temperature.

so, actually comparing core temp. to tcasemax would be incorrect.

you should compare tcase to tcasemax

again, tcase is physically the temperature either on the top of the die or on top of the IHS.

To convert, junction temperature (tjunction or core temp) to case temperature (tcase) you would need to know the thermal resistance of the die (silicon) and the thickness of the die if tcase is defined as the temp on top of the die. Or additionally, you would need to know the thickness of TIM, thermal resistance of TIM, thickness of IHS, and thermal resistance of IHS if tcase is defined as top of the IHS.

Max junction temperature is defined by foundry process only. Wafer subcons like TSMC defined this to be either 90C, 110C, 125C or 150C depending on different process. This is unknown for Intel or AMD since they fab their own wafers....

[aiya]

here's mine



i have a opteron 165 and it seems to show core #2 temp in speedfan 4.29.

[Ychocky]

Lovely little app. Thanks for your contribution, keep up the great work!

[martin_ezz]

Hello,

i'm just wondering why in most cases Core temps (SpeedFan) are higher than Cpu tems???

[Lolento]

Hello,

i'm just wondering why in most cases Core temps (SpeedFan) are higher than Cpu tems???

Your conventional motherboard monitoring programs should be reporting the CASE TEMPERATURE (Tcase) as the CPU temperature. Again, depending how it is defined, it can be the maximum temperature at the top of the die (dead center temperature of the die) or the maximum temperature at the top of the IHS (dead center temperature of the IHS). The key point about Tcase is that it is can be physically validated by a temperature measuring tool (ie thermal couple).

The core temperature as defined in the original post here should be the JUNCTION TEMPERATURE of the die. Compared the CASE TEMPERATURE, JUNCTION TEMPERATURE is less useful to the average system builder or consumer because it cannot be measured physically without destroying the part.

Since Intel and AMD are not system builders (end products are ICs), they defined their products' design limit with Tcasemax (because Tcase can be measured by the customer). This is an industry standard way of reporting thermal limitation of ICs. Of course, they CAN also define Tjunctionmax since Tcase is derived from Tjunction, but customers who buy these ICs would not be able to verify the actual Tjunction. The ability to verify Tcase is very important to system builders (I don't mean average system builders but Dell, HP...) because they need verify the thermal design limit of the system.

Tjunctionmax is usually defined by wafer foundries based on the process technology of the wafer. For example, 0.35um process would have a Tjunctionmax of 150C, 0.13um process would have Tjunctionmax of 125C (don't quote me on this; just for example)...etc And you can see this in Nvidia graphic driver where they set the max core temperature to 125C.

Thus, if you know the Tjunctionmax of the CPUs, then knowing the core temps would be useful.

Or if you know (for example if Tcase is defined as top center of the die) the die thickness and the thermal resistance of the die per thickness and power draw of the die, Tcase can be calculated from the core temperature.

For the newer Flip Chip type IC modules (basically all CPUs and GPUs nowadays), the core temperature would be close the case temperature but they would not be identical since the thermal resistance of bare silicon has to be taken into account.

[dsl]

Under dry ice

[Cybertronic]

Nice program, thanks!

Would it be possible to add an icon to the program please?

[Omarko]

so how does this work with the new Intel Core duo CPUs ?

are the readings we are getting accurate at all ? there seem to be too many discussions at the moment about which program is accurate ...

[Lolento]

so how does this work with the new Intel Core duo CPUs ?

are the readings we are getting accurate at all ? there seem to be too many discussions at the moment about which program is accurate ...

If you're getting the results of the core temperature, you cannot verify it. You just have to take Intel or AMD's word for it.

I'm just pointing out that in this program, the core temperature is compared to Tcasemax which is comparing two different things. I'm not trying to discourage the author of this program, I think the program it is great, but I want to point the inaccurate way of how this program report results so everyone is aware and may be the author will make an update.

[bigKr33]

Holy ****!! i'm only doing 23-25c on core 1 and 31 on core 2. Smart guardian is showing 33-34c. Are you sure this is accurate stuff? MY room temp is 25c, so theres no way water cooling can dip below room temp.

[bigKr33]

Seems everyone's speedfan reads almost exactly the same as this program, but mine is completely off.

[Dragoran]

If you're getting the results of the core temperature, you cannot verify it. You just have to take Intel or AMD's word for it.

I'm just pointing out that in this program, the core temperature is compared to Tcasemax which is comparing two different things. I'm not trying to discourage the author of this program, I think the program it is great, but I want to point the inaccurate way of how this program report results so everyone is aware and may be the author will make an update.

I have written a similar programm for linux...
you only read a value from a register and calculte the temp ( regvalue - 49), so the programm only reports what your cpu provides (in case of amd, I don't have a core2 duo ; and also havn't read the core 2 dup docs)

[Praz]

Since we are talking about inaccuracies we need to discuss the thermal diode. AMD states in their white paper that because of the location of the diode it may read up to 4C cooler then the junction temperature. Then they go on to state that the accuracy of the diode is +/- 10C. They also say that these values do not include any inaccuracies of the sensor on the motherboard. AMD also cautions manufacturers not to use the diode temperature for thermal design calculations.

None of us know how accurate the thermal sensor is built into the processor that Core Temp is reading. What we do know is that the temperature reported by the bios and programs that we have used in the past is at best a general indication of what the actual temperature may be. This information is nor speculation. It's presented by AMD in their design papers.

[Lolento]

Since we are talking about inaccuracies we need to discuss the thermal diode. AMD states in their white paper that because of the location of the diode it may read up to 4C cooler then the junction temperature. Then they go on to state that the accuracy of the diode is +/- 10C. They also say that these values do not include any inaccuracies of the sensor on the motherboard. AMD also cautions manufacturers not to use the diode temperature for thermal design calculations.

None of us know how accurate the thermal sensor is built into the processor that Core Temp is reading. What we do know is that the temperature reported by the bios and programs that we have used in the past is at best a general indication of what the actual temperature may be. This information is nor speculation. It's presented by AMD in their design papers.

This is exactly where I'm getting at. System builders will not rely on reports of junction temperature because it cannot be validated.

[Praz]

Junction temperature can be verified. AMD outlines the procedure in one of their papers. They state the temperature reading will be within 2C of the junction temp. Manufacturers are told not to validate their systems using the thermal diode. The same diode that the bios, MBM and similar programs rely on that we have been using.

[Dave_Sz]

Your conventional motherboard monitoring programs should be reporting the CASE TEMPERATURE (Tcase) as the CPU temperature. Again, depending how it is defined, it can be the maximum temperature at the top of the die (dead center temperature of the die) or the maximum temperature at the top of the IHS (dead center temperature of the IHS). The key point about Tcase is that it is can be physically validated by a temperature measuring tool (ie thermal couple).

what's a thermal couple???

[Lolento]

Junction temperature can be verified. AMD outlines the procedure in one of their papers. They state the temperature reading will be within 2C of the junction temp. Manufacturers are told not to validate their systems using the thermal diode. The same diode that the bios, MBM and similar programs rely on that we have been using.

Junction temperature can be verified. But you have to destroy the chip by drilling a hole from the pin side of the cpu to place the thermal couple.

Case temperature can be verified without any destructive technique.

[Praz]

AMD tells how to measure the junction temperature indirectly. Following their guidelines they claim the readings will be within 2C of actual junction temperature.

Two things we all know for sure. Core Temp is only as accurate as the thermal sensor. No where has any info been reported from AMD as to the accuracy of the sensor. But the info released from AMD states the diode can read as much as 6C high to 14C low. And as AMD says this does not take into account any inaccuracy of the motherboard sensor or the software being used.

[Lolento]

AMD tells how to measure the junction temperature indirectly. Following their guidelines they claim the readings will be within 2C of actual junction temperature.

Two things we all know for sure. Core Temp is only as accurate as the thermal sensor. No where has any info been reported from AMD as to the accuracy of the sensor. But the info released from AMD states the diode can read as much as 6C high to 14C low. And as AMD says this does not take into account any inaccuracy of the motherboard sensor or the software being used.

yes, the indirect method IS measuring the CASE TEMPERATURE, and then reverse derive the JUNCTION TEMPERATURE of the chip. But knowing the JUNCTION TEMPERATURE would not help you since you don't know what is the TjunctionMax.

Any time you measure the surface temperature of the chip, you are measuring the CASE TEMPERATURE.

Again, the CASE TEMPERATURE can be defined differently (however, always on the surface of the IC package) but the easiest definitions are either top of die or top of IHS. This is because these locations make it easy to calculate the JUNCTION TEMPERATURE.

On the other hand, if you define the CASE TEMPERATURE on the bottom of the IC package, it will be more difficult to calculate the JUNCTION TEMPERATURE because new IC packages use organic substrate instead of ceramic (like in the old days). The thermal resistance of organic substrate is hard to characterize due to asymetry in substrate design (copper thickness versus dielectric thickness makes a big difference). On the other hand, older ceramic substrate is much easier to characterize due to tungsten thickness and ceramic thickness have little effect on thermal resistance.

Anyhow, since Tcasemax is given for the AMD chips. You should just follow AMD's direction to measure Tcase. Knowing Tjunction does not tell you anything.

If you read AMD's paper, they assume you understand what is CASE TEMPERATURE and what is JUNCTION TEMPERATURE. So before you read them, read a book on IC thermal management.

I highly suggest you understand these meanings before you read any thermal management papers. Aside from Intel and AMD, other design houses use ThetaJA, ThetaJC, PsiJT...to define thermal design limits. You will not know how to build a reliable system if you don't know what these mean.

Definitions:

Case Temperature (not the pc case temperature) is the surface temperature of the IC package. This is usually defined as the top surface temperature of the IC package. The limitation of case temperature is defined from the maximum junction temperature allowed and the configuration of the IC package.

Junction temperature is the temperature on the active circuitry of the die (...how do you physically measure this without destructive technique once the die is packaged in flip chip form???). The limitation of junction temperature is defined by the wafer techonlogy.