Intel TAT / CoreTemp / IDCC all different temperatures....

[unclewebb]

wow, had no idea my thread was still alive and kicking,

The mystery of the absolute temperature of a Core 2 Duo has still not been solved.

I had a look at your results on the first page of this thread and would have to disagree with you. I believe your processor has a TjMax of 100C and you really were running it at over 90C.

These are just numbers and nothing to get too worried about. I've found that as long as your Orthos stable your C2D will be fine. I ran Orthos for half an hour one day and adjusted the CPU fan to keep it at 3 degrees from TjMax. In my case that was likely 97C and my computer was rock solid stable.

Yours is one of the few processors that I've seen that TAT is reading properly as TjMax=100C. Would it be possible for you or anyone that has a processor that TAT shows numbers above 85C to do a RDMSR on location 0xEE and post your results.

Refer to this post to understand what I'm talking about:
http://www.xtremesystems.org/forums/...&postcount=139

My C2D does not start to throttle until DTS=2. With a TjMax=100C that would be an absolute temperature of 98C. Your test showing your processor running at 94C and not throttling makes perfect sense. It wasn't hot enough yet.

[Bail_w]

Look at this.

[Coffee29]

Thank you for the MSR infor unclewebb. I knew the address of the MSR but was not sure how to really use it. For those people that have problems getting coretemp 95 to run without crashing your system, this is a solution to getting your DTS.

[RealTelstar]

Look at this.

I tend to believe to coretemp 0.95.

Also, I think newer C2D have 100° tcc (tjunction). I'd like to hear from people with supposedly ~100°C tjunction which week is their c2d.

[unclewebb]

Also, I think newer C2D have 100° tcc (tjunction).

I don't think TjMax=85C applies to any of the Core family of processors, mobile or desktop. I think it came about because a programmer or two misinterpreted bit 30 at location 0xEE and the misinformation spread from there.

My E6400 is one of the early ones which CPUz shows as revision B2 and is based on the original Conroe core.

E6400 - SL9S9
Batch#: L633A988
Pack Date: 12/05/06 ( May 12, 2006 )

[kwalker]

I don't think TjMax=85C applies to any of the Core family of processors, mobile or desktop. I think it came about because a programmer or two misinterpreted bit 30 at location 0xEE and the misinformation spread from there.

My E6400 is one of the early ones which CPUz shows as revision B2 and is based on the original Conroe core.

E6400 - SL9S9
Batch#: L633A988
Pack Date: 12/05/06 ( May 12, 2006 )

You may be right
no one really knows at what point TCC activation occurs for actual core temp calculations read from the DTS.
with the introduction of the Q6600 and a few other newer processors the elimation of the Thermal diode takes away another factor (Tcase)
the temperatures are read directly from the DTS and discrepencys are surfacing in the software readings we thought were reliable.
I will keep following this thread
perhaps some light can be shed on this subject in time.

[Prefixxx]

Just FYI

Send an email to Intel asking about temps and this what I get back.


Hello xxxxx,

Thank you for contacting Intel(R) Technical Support.

Please pay attention to the CPU temperature.
You do not need to pay attention to the core temperature.

If the CPU temperature is lower than 60.1 everything should be OK.

Sincerely,

Axxxx S.
Intel(R) Technical Support

[RealTelstar]

doh!
we've been ignoring the "cpu" temp and looking at the cores so far...

[Elisha]

Just FYI

Send an email to Intel asking about temps and this what I get back.




Please pay attention to the CPU temperature.
You do not need to pay attention to the core temperature.

I agree there....because when i test Orthos and the CPU temp goes above 70.....Orthos fails nomatter what the Core Temp is....Core temp has gone upwards to 79 and has been stable.
but when i open my case side panel.....cpu temps do not exceed 63 degrees. with it closed it get to 70 and above and i have placed a fan everywhere i can think a sensor was located since i thought speedfan was actually reading the NB or PWN temp as CPU temps. but nomatter where i place the fan....the temps do not decrease.

[fgw]

Just FYI

Send an email to Intel asking about temps and this what I get back.


Hello xxxxx,

Thank you for contacting Intel(R) Technical Support.

Please pay attention to the CPU temperature.
You do not need to pay attention to the core temperature.

If the CPU temperature is lower than 60.1 everything should be OK.

Sincerely,

Axxxx S.
Intel(R) Technical Support

the question here is: where is this cpu temperature measured?
looking at the figure of 60.1c i assume this intel guy is talking about tcase as documented in the processor specs. besides tcase is different for different processor families, unfortunately no software is showing this temp anywhere. the recommended method of measuring tcase is by putting a temp probe on the ihs. from looking at intel docs, this is a very tricky task and i doubt many people out here will do this ...
the cpu temperature displayed in bios or read by any other software is simply the temperature measured via the thermal diode and calibrated by some unknown factor. this reading would vary from motherboard manufacturer to motherboard manufacturer and even from one bios rev to the other. don't assume to reading tcase here!

[unclewebb]

I finally got around to doing some more testing today. I'm not quite sure what I've proven besides human stupidity has no bounds but I did end up with lots of interesting data.

I purchased a Fluke IR thermometer recently and a crazy guy at the [H]ard forum jokingly suggested that to get an accurate core temperature reading I should pull off the heatsink and the IHS so I can get a direct shot at the processor. That sounded like an interesting idea so I decided to be the first on my block with a naked Core 2 Duo. I left the IHS on but after it booted up I pulled off the OEM heatsink and fan while it was running. Two strips of masking tape over top of the processor takes away the shine and helps ensure repeatable temperatures from the IR thermometer.



To keep the processor from getting too hot I used an 80mm high speed hand held fan and had no trouble keeping the processor running in the 60C to 70C range.



CoreTemp 0.95 which uses a TjMax=85C for my E6400 was in agreement with the IR readings I was getting at 60C, 70C, 80C and beyond.

To explore when the throttle point kicks in I decided to put the hand held fan down and take pictures as the processor warmed up. Here was the last picture I took before things started to get serious.



When the IR reported my cores getting near 85C, CoreTemp started spitting out these errors every couple of seconds.



I thought for sure a crash was imminent but it kept running as Intel Thermal Monitoring kicked in trying to regain control of this processor by throttling it.



We need a time out to explain the temperature numbers being displayed by these programs.

The DTS is a 7 bit number. What seems to be happening is DTS gets to zero and then when the processor gets 1 degree hotter the value in DTS goes from 000 0000 to 111 1111 binary. That is equivalent to 127. CoreTemp shows with a question mark that there is something wrong here and then reports the value in DTS directly.

CoreTemp -> DTS -> Real_Temp
84 -> 1 -> 84
85 -> 0 -> 85
127 -> 127 -> 86
126 -> 126 -> 87
125 -> 125 -> 88

SpeedFan and TAT work differently. They continue to use the standard formula of Real_Temp = TjMax - DTS so in this case they display:
Temp = 85 - 127 = -42

They report negative numbers starting at -42 and then as the processor continues to heat up the DTS continues to decrease just like usual.

The CoreTemp log file looks interesting.



Core0 maxes out at a DTS value of 116 which, if my calculations above are correct, is actually 97C. Core1 seems to be displayed at a constant 85C reading when temperatures are out of control.

In the following graph, SpeedFan shows after it gets to 85C there is a sharp drop to -42 and then it continues to increase from there. Before my processor caught on fire I decided to put the hand held fan back on it and was able to regain control of this overheating processor.



After taking this processor to hell and back I learned something that I already knew.
The Core 2 Duo is a hell of a processor.

[Elisha]

very nice find.....so what is the CPU temp measurement in Speedfan?
i find that Orthos fails when that hits 70 nomatter what the Core0 and Core1 temps are and nomatter what the cpu voltage is at 3.6Ghz.

[unclewebb]

I've gone back and forth but after seeing the results I would have to say that my TjMax really is 85C. I had no trouble stabilizing my processor at various temperatures from 60C to 80C and each time the IR thermometer reported a temperature within 1C of what CoreTemp with an 85C TjMax was displaying.

In my case, if TjMax was 100C like I had thought then that would have to mean that the individual cores are running 15C hotter than the IHS which is touching them. Without a heatsink attached, I can't believe that there is 15C of heat dissipation happening in only the thickness of the IHS.

The masking tape used on the processor for testing has previously proven to provide no difference in the temperature of the object underneath it. With my test computer shut down and unplugged for the last couple of hours, the IR thermometer reports the CPU to be exactly equal to the ambient temperature which validates my testing procedure.

If the above is all true and DTS = 85C for my processor then I would also have to conclude that the DTS sensor is not accurate at lower temperatures. Intel designed the DTS and calibrated it to accurately control the Thermal Monitoring functions when a processor gets near the TjMax point. It shouldn't be too surprising if it's not very accurate at reporting very low idle temperatures.

That also explains why Intel refuses to document an absolute core temperature for a Core 2 Duo. Any calculation based on the DTS wouldn't be accurate for the entire operating temperature range of the processor.

Most importantly, the XS guys seem to be 100% right, again! Trust the DTS and unless you want to invest in an IR gun, don't put too much faith in any absolute temperatures for the Core 2 Duo.

I'll try to post some temps of the copper plate on my new Tuniq when I have that up and running but I think I've learned all that can be learned when it comes to C2D temperatures.

[unclewebb]

Elisha: On my E6400 I have run Orthos stable for half an hour at 3200 MHz with my processor within a degree of starting to throttle. At 3400 MHz the absolute core temperature can't go beyond 75C while running Orthos before it crashes within seconds but it can run Orthos stable for hours at 3400 MHz as long as the core temperature stays down in the low 70C range.

If you are at or near your maximum MHz overclock then you will need to run your processor at a lower core temperature to maintain stability. The CPU temperature reported by SpeedFan is based on an inaccurate diode reading. Intel designed the DTS because the diode based temperature readings weren't accurate enough so ignore the SpeedFan reported CPU temperature.

[joebuffalo]

How did you come to the conclusion that DTS was not as accurate at lower temperatures?

[unclewebb]

How did you come to the conclusion that DTS was not as accurate at lower temperatures?

Here is the [H]ard forum where I've been discussing this issue:
http://www.hardforum.com/showpost.ph...&postcount=165

I did a previous test with just the OEM heatsink attached and then I used my 80 mm hand held fan to get my core temp down. This allowed me to get in there and measure the temperature of the copper core inside the heatsink without a fan being in the way.

Based on a TjMax=85C, CoreTemp was reporting a very steady 21C while the IR thermometer was reporting that the center copper cone of the heatsink was at 27C. That's impossible. The heatsink can't be warmer than the processor it's cooling.

I thought that must mean that TjMax for my processor is 100C but that doesn't seem possible after my test above.

From the data gathered my only conclusion is that the DTS can't possibly be 100% accurate from idle temps to DTS=0 temps. I'm really open to any other possible explanation for what I've found.

[unclewebb]

Just for fun I took my final graph yesterday and decided to use Photoshop to cut out the part of the graph when SpeedFan started reporting negative numbers. I moved that piece of the graph straight up and it fit into the upper part of the graph perfectly.



This seems to show that both cores continued to run even when the core temperature was up over 90C.

[RealTelstar]

I think i want to invest in a IR thermometer for fun
There are other components besides the cpu that I want to check (ram and my hot scsi controller on top). what do u suggest under $50 and available in EU?

[Coffee29]

Unclewebb...just a thought.

If I understand your logic you seem to have argued the following: your previous testing suggested that a tjmax of 85 could not be right and you thus assumed it to be 100. However, your more recent testing is in favor of the 85 over the 100, thus you have now concluded that your tjmax may in fact be 85. In order to remedy the inconsistency of the former testing you conclude that DTS may not be accurate at lower temps.

Correct me if I am wrong about this summation, however, if I have stated the basic argument correctly, isn't a more plausable explanation simply that the tjmax you are looking for does not exist? Possibly DTS is accurate all of the time and the reason that you are getting conflicting results is not due to invalid DTS readings at lower temps, but instead because you are maintaining that your processor uses a tjmax variable of 85 or 100?

[unclewebb]

Every C2D has to have some fixed TjMax value associated with it where the processor begins throttling. It might be 85C or 100C or some number in between but it has to be some number. The problem I'm having is coming up with one number that explains what I'm seeing at both idle and at high temperatures.

Assuming a TjMax of 85C for my E6400 results in absolute temperatures reported by CoreTemp that are within 1 degree of what my IR thermometer is reading pointed directly at and hovering just above the cores.

Yesterday I had no idea what I was in for and the core temp went sky high as I was busy taking pictures and saving screen shots. I put the camera aside today and just concentrated on comparing the readings from CoreTemp to the IR thermometer. By adjusting the distance and angle of the fan to the cpu, I was able to run at a variety of temperatures between 48C and 70C without a lot of variation. I had plenty of time to slowly position the IR thermometer directly over the cores and obtain readings while the core temperatures remained steady.

The IR readings were generally the same as CoreTemp or within 1C at any point from 48C to 70C.

I installed a Tuniq Tower this evening and so far I've just been running it at low MHz and low volts. The ambient temperature is 23C and CoreTemp using an 85C TjMax is reporting 19C. That's not possible. Worse than this is that the copper plate on the bottom of the Tuniq which presses against the IHS is at 29C. For this to be possible, CoreTemp needs to be reporting at least 10C higher so that implies TjMax needs to be 95C for the idle temps to make sense.

TjMax = 95C is impossible at higher temperatures over 50C. How could two cores that are both at 60C only be heating the IHS up to 50C? The IHS is designed so that it is physically contacting both cores and the IR thermometer is located within 1 mm or 2 mm of the IHS.

That's the dilemma. There's no single TjMax number that covers both situations at idle and at high temps. The only thing I can conclude from this is that the DTS is calibrated and very accurate at the trip point but isn't as accurate at very low idle temps.

SuperKeijo on the [H]ard forums has also removed his heatsink and is experiencing this same dilemma during testing of his E4300 with an IR thermometer. Hopefully he'll have some more results this weekend.

RealTelstar: Extech makes IR thermometers and are available new on EBay for about $50 to $60. I think an IR-205 would be sufficient for this kind of testing.

[unclewebb]

....so what is the CPU temp measurement in Speedfan?

Here is some proof that the CPU Temperature number based on the on board diode is a completely meaningless number.



With an ambient temperature of 22C the bios displays 13C when it reads this diode and SpeedFan reports 12C which are beyond meaningless. Turn it off and ignore it because inaccurate sensors can't provide you with any reliable information that can be used to determine anything.

[joebuffalo]

Before I jump in on this discussion again, I want to state the point of this thread one more time for any new readers who have not taken the time to read all the posts:

Absolute temperature are worthless. All you need to know is DTS. Stop worrying about anything else!

unlewebb, your assumption that you are reading the core temperature as measured by DTS when you use your IR thermometer on the heatsink or even the IHS is just wrong. There is DEFINITELY a thermal gradient that you are simply ignoring. The reason Intel implemented multiple digital temperature sensors is because they knew that there are multiple hot-spots which could not be captured by a singe sensor.

From http://documents.irevues.inist.fr/bi...79/1/TMI23.pdf
(emphasis is mine):

"It can be seen that large temperature gradients exist on the
die. It also can be noted that some workloads display high
temperature gradients while other have no offset. Thermal
control algorithms need to prevent the hot spot from
exceeding the max temperature specification. It is possible
to mitigate the temperature difference by applying a fixed
offset to the diode reading. This obviously is a non
optimal solution as the workloads with low offset will be
panelized by the unnecessary temperature offset. The use
of digital thermometer provides improved temperature
reading, enables higher CPU performance within thermal
limitations and improves reliability."

If there are larger temperature gradients ON THE DIE surely they exist across the TIM, IHS, more TIM, heatsink, etc.

[TheMafioso]

SuperKeijo on the [H]ard forums has also removed his heatsink and is experiencing this same dilemma during testing of his E4300 with an IR thermometer. Hopefully he'll have some more results this weekend.

What does his tests about the e4300 conclude, has he been able to figure out if it has TjMax of 85C like your e6400 or is TjMax 100C as taken by tat/coretemp 0.95

(I looked for it in the [H] thread, but couldn't find/missed his results)

[kwalker]

Before I jump in on this discussion again, I want to state the point of this thread one more time for any new readers who have not taken the time to read all the posts:

Absolute temperature are worthless. All you need to know is DTS. Stop worrying about anything else!

unlewebb, your assumption that you are reading the core temperature as measured by DTS when you use your IR thermometer on the heatsink or even the IHS is just wrong. There is DEFINITELY a thermal gradient that you are simply ignoring. The reason Intel implemented multiple digital temperature sensors is because they knew that there are multiple hot-spots which could not be captured by a singe sensor.

From http://documents.irevues.inist.fr/bi...79/1/TMI23.pdf
(emphasis is mine):

"It can be seen that large temperature gradients exist on the
die. It also can be noted that some workloads display high
temperature gradients while other have no offset. Thermal
control algorithms need to prevent the hot spot from
exceeding the max temperature specification. It is possible
to mitigate the temperature difference by applying a fixed
offset to the diode reading. This obviously is a non
optimal solution as the workloads with low offset will be
panelized by the unnecessary temperature offset. The use
of digital thermometer provides improved temperature
reading, enables higher CPU performance within thermal
limitations and improves reliability."

If there are larger temperature gradients ON THE DIE surely they exist across the TIM, IHS, more TIM, heatsink, etc.

The best thing about the digital temperature sensors is that Intel has taken the guess work out of the equation in thermal managemnet.
I do agree that the thread needs to stay on track but I was waiting for the digital pictures of a smoldering processor or better yet
Flames

[unclewebb]

Absolute temperature are worthless. All you need to know is DTS. Stop worrying about anything else!

With TjMax and absolute core temperatures still undocumented by Intel, that statement remains true.

unlewebb, your assumption that you are reading the core temperature as measured by DTS when you use your IR thermometer on the heatsink or even the IHS is just wrong. There is DEFINITELY a thermal gradient that you are simply ignoring.

Here I disagree. I definitely haven't ignored the potential gradients and have even done some testing of this. Gradients exist when an application works one part of the core while other parts of the same core are mostly idle. For example this will happen if a program is working on a large calculation that involves mostly floating point operations. The floating point area of the core is going to be running at a hotter temperature than other surrounding parts of that same core. Here's how I tried to determine how big this gradient might be.

The process that is known to create the most amount of heat in a Core 2 Duo is MeromMaxPowerVer0p3.exe which is the name of the process that TAT runs when you tell it to run a load. I decided to run this process on one core while the second core was idle and then I switched the Affinity of this process while it was running to get it to immediately switch over and run on the second core so the first core now became idle. The difference in temperature between the two cores averaged 6C with the OEM heatsink and fan. With a better cooling solution the difference would likely be slightly less.



To me that represents a worst case scenario. Having one core running the most demanding program at full load heats up the other totally separate core to within 6C even though the second core is idle doing nothing. That surprised me because I was expecting a larger gradient to exist. This shows that the heat of each individual core is being shared with the other core and more importantly the graph also shows how quickly a core will decrease in temperature to its surroundings when it becomes idle. CoreTemp logging showed that this took less than a second for each core to stabilize at its new temperature.

With virtually zero load on both cores, with the heatsink and TIM removed and being cooled with a hand held high speed fan, the gradient between the two cores drops to zero. Both cores report that they are operating at the exact same temperature. Gradients are caused when part of a core is under load while other parts of the same core are mostly idle but if the entire core is idle than the gradients do not exist or are so small that they can be ignored.

The hot spots shown in the Intel diagrams are for a processor that is working but those hot spots don't apply to a processor that is idle and allowed to sit and stabilize at a fixed temperature.

A consumer hand held IR thermometer is not accurate enough to acquire rapidly changing temperature data from a processor that is under load but at idle, it's a different story.

The temperature of the cores at idle becomes the same and they also heat up the center of the IHS to a similar temperature. The round opening in the IR thermometer is approximately 1 square inch so it is taking data from both cores and the space between them and averaging it out to display one temperature number. My opinion is that at idle, the temperature gradient between the two individual cores and the IHS where I'm taking readings from becomes insignificant and certainly less significant than the 1% sampling error of the IR thermometer.