No Accurate Way of Measuring Temp for C2D and C2Q Series?
- Thread starter geokilla
- Start date
Most temp programs measure the diff. between the tjunction and current temp. They then take their tjunction numbers and subtract that number getting your temp. Basically tjunction is when your cpu throttles, so to find out how far from throttling you are go into the options menu for coretemp (probably for real temp too, not sure about speedfan though, I don't really use it) and select delta Tjunct. Just stay 15 to 20 degrees away from that and your cpu life shouldn't be effected.
Guess reading comprehension isn't your strong suit, eh?
What's been said and is being said, by Intel, Anandtech, and others....is that NO ONE, outside of Intel (and they're not telling!), knows what the tjunction number really is......most of these programs are basing their tj number on a guess, or taking what's known about the mobile processors and trying to correlate that info to the desktop processors.
So, in all actuality, the tjunction that Speedfan, Coretemp, and others are using could be waaaay off......like even 15C, maybe more. So in essence, the info that some programs are giving about temps is essentially useless and can only give you a trend for where your temps are headed....you don't even know if you are within 10-15C of throttling...even if you choose delta tjunction, that's just as useless.
- Oct 14, 2006
- 2,012
- 3
- 81
So no one cares about this?
Like what ginger said, the fact that Intel didn't publish the tjunction values are pratically making Core Temp and Real Temp and whatever program you use to monitor the CPU temps useless! For all we know, the Tjunction could be something like 80C while Core Temp says 100C or 85C.
Like what ginger said, the fact that Intel didn't publish the tjunction values are pratically making Core Temp and Real Temp and whatever program you use to monitor the CPU temps useless! For all we know, the Tjunction could be something like 80C while Core Temp says 100C or 85C.
FYI
http://www.tomshardware.com/fo...quad-temperature-guide
I've seen it said before that for parts with the DTS which is relative to TJMAX maybe it is better to NOT try to correct them to 'absolute temperautre' but to just rely on the delta temperature. After all for a maximum overclock your LIMIT is always going to be temperatures of 0 delta = Maximum allowed core temp - Actual Core temp.
Whatever Intel considers "Maximum allowed core temp" for your part is irrelevant sort of because if you get close to hitting that, your CPU will start thermal throttling and degrading rapidly electrically and in performance etc. and your system may shut itself down etc.
Of course it's always better to have a higher DELTA between the maximum allowed safe operating temperature and the actual temperature, so if you're tuning the system, aim for the biggest delta you can generate, that'll always mean you're running COOLER than with less deltas.
Given that the actual temperature is just a nice thing to know if you're trying to match it against your room temperature, PWM temperature, motherboard temperature, etc. but you can always just look at the Delta TJMAX versus the other ones to see where everything is in a relative sense.
That all said, IMHO it is stupid that they didn't give you a way to get ACTUAL ABSOLUTE temperature from the chip AS WELL as the more important delta TJMAX... $300+ CPUs with the power of supercomputers as of 20 years ago and they can't even tell time or temperature well yet like a $15 wrist watch... brilliant.
http://www.tomshardware.com/fo...quad-temperature-guide
I've seen it said before that for parts with the DTS which is relative to TJMAX maybe it is better to NOT try to correct them to 'absolute temperautre' but to just rely on the delta temperature. After all for a maximum overclock your LIMIT is always going to be temperatures of 0 delta = Maximum allowed core temp - Actual Core temp.
Whatever Intel considers "Maximum allowed core temp" for your part is irrelevant sort of because if you get close to hitting that, your CPU will start thermal throttling and degrading rapidly electrically and in performance etc. and your system may shut itself down etc.
Of course it's always better to have a higher DELTA between the maximum allowed safe operating temperature and the actual temperature, so if you're tuning the system, aim for the biggest delta you can generate, that'll always mean you're running COOLER than with less deltas.
Given that the actual temperature is just a nice thing to know if you're trying to match it against your room temperature, PWM temperature, motherboard temperature, etc. but you can always just look at the Delta TJMAX versus the other ones to see where everything is in a relative sense.
That all said, IMHO it is stupid that they didn't give you a way to get ACTUAL ABSOLUTE temperature from the chip AS WELL as the more important delta TJMAX... $300+ CPUs with the power of supercomputers as of 20 years ago and they can't even tell time or temperature well yet like a $15 wrist watch... brilliant.
It isn't TOTALLY up in the air as to the proper correlations either..some of the more
extreme overclockers do things like cool their CPUs with water or even phase change or cryogenic coolers.
Some even remove the IHS and put the heatsink right on the die.
In those cases you can be basically 100% sure what the heatsink's temperature is because you know your water temperature or the heastink block temperature pretty well, as well as have a good idea of what your thermal resistance is from your coolant to the top of the CPU die.
Certainly in such cases it is true when the PC is LIGHTLY loaded or IDLE the heat the CPU will be generating will be pretty insignificant compared to the cooling capacity of those kinds of high end coolers, especially for a 45nm CPU.
Therefore if you look at the DTS temperature on a VERY WELL cooled idle CPU with a KNOWN heatsink temperature and very little thermal resistance between the core and the heatsink, you are going to be pretty sure within like 0 - 5 degrees C that your absolute core temperature is the same as your heatsink/coolant temperature, and hence have some reference point between real world temperatures and measured relative temperatures.
Similarly again if you mod your IHS and put a thermocouple between the HS and IHS you'll know precisely what the IHS / CPU case temperature is in that area. Then you measure the thermal diode or DTS sensor and see what it says at a variety of CPU loads.. at low CPU loads the CPU core temperature will effectively asymptotically approach equality with the known IHS/heatsink temperature, and you have a point of correlation.
This is basically what the Intel thermal management heatsink design whitepaper tells you to do, I've read it, though haven't bothered to make any thermocouple heatsinks yet.
Anyway from people doing these kinds of analyses you can pretty much narrow down what the correlation is between DTS and thermal diode temperatures and real world temperatures within a few degrees C assuming that Intel doesn't change the way their chips work too much with different steppings / lots, it'll be about the same for everyone.
Still kind of pathetic that they don't just document the correlation.
extreme overclockers do things like cool their CPUs with water or even phase change or cryogenic coolers.
Some even remove the IHS and put the heatsink right on the die.
In those cases you can be basically 100% sure what the heatsink's temperature is because you know your water temperature or the heastink block temperature pretty well, as well as have a good idea of what your thermal resistance is from your coolant to the top of the CPU die.
Certainly in such cases it is true when the PC is LIGHTLY loaded or IDLE the heat the CPU will be generating will be pretty insignificant compared to the cooling capacity of those kinds of high end coolers, especially for a 45nm CPU.
Therefore if you look at the DTS temperature on a VERY WELL cooled idle CPU with a KNOWN heatsink temperature and very little thermal resistance between the core and the heatsink, you are going to be pretty sure within like 0 - 5 degrees C that your absolute core temperature is the same as your heatsink/coolant temperature, and hence have some reference point between real world temperatures and measured relative temperatures.
Similarly again if you mod your IHS and put a thermocouple between the HS and IHS you'll know precisely what the IHS / CPU case temperature is in that area. Then you measure the thermal diode or DTS sensor and see what it says at a variety of CPU loads.. at low CPU loads the CPU core temperature will effectively asymptotically approach equality with the known IHS/heatsink temperature, and you have a point of correlation.
This is basically what the Intel thermal management heatsink design whitepaper tells you to do, I've read it, though haven't bothered to make any thermocouple heatsinks yet.
Anyway from people doing these kinds of analyses you can pretty much narrow down what the correlation is between DTS and thermal diode temperatures and real world temperatures within a few degrees C assuming that Intel doesn't change the way their chips work too much with different steppings / lots, it'll be about the same for everyone.
Still kind of pathetic that they don't just document the correlation.
Well Real Temp 2.21 is saying I'm idling at 38C @ 8x500 w/ 1.375vcore. I touch the HSF it's cold room temp even right on the base that's touching the CPU. It's not even int he least bit warm. So is it really that temp at the CPU? I have PLENTY of cooling in this case.
This is one advantage of having an nvidia 780i, it has ESA which gives more accurate measurements from temp and lots of stuff.
Anandtech's article: http://www.anandtech.com/showdoc.aspx?i=3145&p=3
Anandtech's article: http://www.anandtech.com/showdoc.aspx?i=3145&p=3
The "touch" method is a poor indicator of how warm the CPU is at the transistor level far away from even the IHS.
The only way your HSF would be the same temp as the transistors is if the thermal conduction were infinity rather than the non-infinite numbers they really are.
This doesn't answer your question, no one can tell you what temp your CPU really is. But given that at least Coretemp/Realtemp can give you a reasonable guide as to what the temps are.
What you seem to be missing here gingerstewart55, is that the value provided by the Digital Thermal Sensor is not the actual temperature of the core. The DTS reports delta Tjunction aka the difference between core temperature and the processor's tJunction.
Then the reporting software like CoreTemp, SpeedFan, etc subtract that delta value from a known (guessed) value for Tjunction.
The issue is that the Tjunction values used by CoreTemp, Speedfan, etc may not be as accurate as once believed, leading to inaccurate core temperature readings.
The solution as pointed out by zach0624 is to rely on delta tJunction rather than computed core temperature.
I have been stating this for over 6 months. For that matter, when people report that their B3's are running so much hotter than a G0, take it for a grain of salt. It is true that the G0's require 15% less power and thus outputs 15% less wattage (Heat), but that wouldn't equate into that large of a temperature difference that is reported by users. The only thing you have is a 'ball park' figure when comparing the same model and stepping. But that isn't good enough to make any real conclusion on. You have far to many variables for temperatures to be accurate at this point and Intel not releasing their TJunction values is just one of many reasons why we don't really know the true temperatures of our C2D/C2Q's.
so when Coretemp/Realtemp report Delta to TJunction Max readings of about 60C, i have 60c more headroom to go?
so the max temp before my cpu throttles is at 95c (or 100c)?
then whats the significance of the thermal spec of 71c?
I'm not disagreeing, but I have to shoot a lot of temps when we do start ups on pumps, compressors and turbines at work, my touch is pretty decent imo. If the HSF at the base isn't even above room temp and I'm stressing the system I'm pretty sure the CPU isn't much hotter. Sure it's not going to be acurate, but it shouldn't be to hot to where it's going to have a melt down. It doesn't get even close to as warm as the bottom of my GFs laptop and she's got one of the newer celeron single cores that thing gets scortching.
i don't get exactly what you're saying
are you saying your sense of touch can tell between cold, room temperature, warm, hot, and scorching?
Well Louiss, i work around a lot of things that get pretty f'in hot heh while running some way to hot to touch, but I've gotten pretty decent at telling temp ranges just by touch when I'm waiting on someone to bring over the laser to shoot the temps. I can tell when somethings 100F(38C) or close to it by touch. There's a difference between 100 and 120 I can tell I guess? I know it sounds ridiculous, but doing it constantly just cause it's an interesting thing to do at work you kind of pick it up? Just like I can look at a bolt or nut and tell you within a 16th of an inch what it is from 3/16ths all the way to 3in. Just things you pick up on the way to make your job easier heh. Anyone who works in my kind of trade can tell you you pick things like that up, got to have an eye micrometer if you want to be taken seriously.
- Oct 14, 2006
- 2,012
- 3
- 81
This is what I would do if I owned a C2D, C2Q. Use CoreTemp/RealTemp to get an IDEA of what my temps are, then leave it be. I doubt there will be a program that can actually show the correct temps unless Intel publishes their Tjunction values for their CPU.
Someone needs to start forcing Intel to give us the Tjunction values for their CPUs, unless the engineers themselves don't know that number either.
Someone needs to start forcing Intel to give us the Tjunction values for their CPUs, unless the engineers themselves don't know that number either.
Intel knows the numbers, they just will not release them. They want us to use cpu temps not core temps. One reason is due to nonlinearity of core temps until delta to TCC is 20-30, thus absolute temps will be inaccurate using core temps until the upper range (1). Another reason is specs are listed for Tcase temps which is more similar to cpu temps than core temps, even though throttling is designed to occur as Tcase max is reached/exceeded (though that ratio varies on too many things and is thus not guaranteed to occur) (2). And intel has told us multiple times if we use core reading, then use them in their accurate form of delta to TCC. Beyond those reasons that intel has stated, only they know of any other issues that prevents them from releasing "tjmax" values. Add on the fact that the nonlinearity increases for 45nm, and I doubt intel will be releasing tjmaxes anytime soon (3).
(1) "PECI counts are roughly linear in relation to temperature changes in degrees Celsius.
For example, a change in PECI count by '1' represents a change in temperature of
approximately 1°C. However, this linearity cannot be guaranteed as the offset below
TCC activation exceeds 20-30 PECI counts."
http://download.intel.com/design/xeon/guides/318611.pdf page 20
(2)"PROCHOT# is designed to assert at or a few degrees higher than maximum TCASE (as
specified by the thermal profile) when dissipating TDP power, and can not be
interpreted as an indication of processor case temperature. This temperature delta
accounts for processor package, lifetime, and manufacturing variations and attempts to
ensure the Thermal Control Circuit is not activated below maximum TCASE when
dissipating TDP power."
http://download.intel.com/design/xeon/guides/318611.pdf page 20
(3) http://www.national.com/appinf...psensors/trutherm.html
though a biased comparison on DTS for 45nm, it illustrates some of the issues reading 45nm transistor/diodes and reporting accurate temps
(1) "PECI counts are roughly linear in relation to temperature changes in degrees Celsius.
For example, a change in PECI count by '1' represents a change in temperature of
approximately 1°C. However, this linearity cannot be guaranteed as the offset below
TCC activation exceeds 20-30 PECI counts."
http://download.intel.com/design/xeon/guides/318611.pdf page 20
(2)"PROCHOT# is designed to assert at or a few degrees higher than maximum TCASE (as
specified by the thermal profile) when dissipating TDP power, and can not be
interpreted as an indication of processor case temperature. This temperature delta
accounts for processor package, lifetime, and manufacturing variations and attempts to
ensure the Thermal Control Circuit is not activated below maximum TCASE when
dissipating TDP power."
http://download.intel.com/design/xeon/guides/318611.pdf page 20
(3) http://www.national.com/appinf...psensors/trutherm.html
though a biased comparison on DTS for 45nm, it illustrates some of the issues reading 45nm transistor/diodes and reporting accurate temps
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter