sportsfanboy,
Zorg and
randomizer,
Allow me to elaborate on
unclewebb's reply.
Let's start with Intel's Thermal Specification, as it applies to desktop processors:
"
maximum case temperature at the maximum Thermal Design Power (TDP) value ... measured at the geometric center on the topside of the processor integrated heat spreader."
Intel's Thermal and Mechanical Design Guidelines describes in great detail how they cut a precision groove into the integrated heat spreader in their labs, in order to embed a highly accurate thermocouple for controlled temperature testing with the stock heat sinc. This is how they accomplished measuring Tcase Max to the tenths of a degree, such as 62.2c for the Q6600 B3 at 105 Watts. At the same time, the Core temperatures are about 25c higher, which illustrates the considerable thermal gradient between the IHS and the DTS, and explains why users are frequently puzzled as to why their cooler only feels warm to the touch while running Prime95 Samll FFT's.
Tjunction Max is calibrated so that Throttling occurs a few degrees higher than the gradient, so the total gradient between Tcase Max IHS temperature and Tjunction Max is typically about 27 to 28c. This is why Tjunction Max becomes just a rounded or "target" figure to the nearest 5c such as 85c or 100c, due to variables such as DTS accuracy, calibration deviations, and sensor "slope error".
rge was recently able to duplicate Intel's lab tests by obtaining the proper type of thermocouples and test equipment, then audaciously drilled holes through the stock heat sincs and integrated heat spreaders of a few sacrificial Core 2 variants. His findings confirmed that the IHS to DTS gradient is as mentioned above, + / - a few degrees.
As the end user has no way to duplicate Intel's lab test setup, and since Intel states that the DTS array was designed for Throttle and Shutdown protection only, we are all instead supposed to be using the Analog Diode CPU temperature, which
is designed for accurate temperature monitoring up and down the scale, and is integrated into the CPU package substrate layers, bottom dead center, for this very purpose. This is why motherboard OEM monitoring utilities, which are included on the driver CD, do not feature Core temperature monitoring, as per agreement with Intel.
I've always found that although BIOS is typically miscalibrated by at least a few degrees, the Analog Diode CPU temperature is consistently accurate and linear
when calibrated, and I've only once seen one "stick". Since
rge was also able to verify that the calibrated Analog Diode CPU temperature is 5c less than the DTS Core temperaure at load, his findings confirm Intel's results shown on page 4, figure 5 in the following document -
http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf - with which
unclewebb,
rge and I are very familiar, and frequently use for reference.
This means that if the Analog Diode CPU temperature is reading 95c, (which is calibrated by BIOS, but we
can measure and re-calibrate), then the DTS Core temperature should be about 100c, (which is calibrated by Intel, but we
can measure and re-calibrate), so the actual IHS temperature should be about 70c, (which we
can't measure). If, for example, we plug in the IHS (Tcase Max) value for the Q6600 G0, then we have the following:
Tcase Max = 71c
Analog Diode = 95c
Tjunction Max = 100c
Tc / Tj Gradient = 29c
When
unclewebb began IR testing with the cooler removed from the IHS, he determined that a Tjunction Max of 95c was appropriate, however, I always thought that he was low by perhaps 2 degrees due to minor factors such as the thermal resistance from the hot spots within the cores through the sensor bonding agent and thermal compound under the IHS, to the masking tape he applied to the top of the IHS to supress IR scatter. Over the past 4 months, I've re-tested 65nm variants and re-verified my original findings for the Guide, as well as having conducted new IR testing and analysis using 3 temperature plots instead of 2, in order to conduct a comparison study of "slope error" and calibrations for the following 65nm and 45nm dual and quad variants:
65nm - E2160 L2, E4500 M0, E6600 B2, E6850 G0, Q6600 G0
45nm - E5300 R0, E7200 M0, E8500 C0, Q9650 E0
The abridged results are that I found for a given processor variant, whether dual or quad, and regardless of whether or not Real Temp used, for example, Tjunction Max 95c or 100c, by using the calibration procedure in the Guide for SpeedFan, (Tcase Load + 5c = Tjunction Load), with which I know you're all familiar, I consistently found that Tjunction Max calibrated values were 97c to 98c, which I believe to be highly accurate, and helps to prove just how very close
unclewebb was. So just for fudge factor, let's round up my findings to 98c and tweak the numbers:
Tcase Max = 71c
Analog Diode = 93c
Tjunction Max = 98c
Tc / Tj Gradient = 27c
Gradient look familiar? Much of
unclewebb's and
rge's recent testing shows that Throttling typically begins a few degrees below Tjunction Max, and that common Tc / Tj gradient values are about 27 to 28c. Now, let's summarize this information and put it into perspective.
(A) We have Intel's IHS lab test temperatures which the end user can't duplicate, but are the temperatures shown for Thermal Specifications in the Processor Spec Finder.
(B) We have the Analog Diode CPU temperature which Intel expects us to use.
(C) We have the DTS Core temperature which Intel does not want us to use.
(D) As experienced overclockers, we all know that hot = unstable, so we don't exceed 25c to 30c Distance to Tjunction Max.
You've all seen me preach on numerous Forum threads:
The temperature shown in Intel's Processor Spec Finder for Thermal Specifications is CPU temperature (Tcase), not Core Temperature (Tjunction), which is a very common misconception among many users.
So the contradiction in terminology becomes obvious; contrary to what I show in the Guide, Tcase and Analog Diode CPU temperature are
not technically synonymous terms. As I know you all appreciate the complexity of this topic, I thought if I introduced yet another definition and deeper concepts into the Guide, it would become too difficult for all but the most advanced users to understand, and as you're all aware, there's been enough heat :lol: about it being too complicated as it is. This is why I state right up front:
... certain strict definitions have been relaxed to simplify concepts ...
As I'm sure you recall,
randomizer, you and I went over this point somewhat recently, and
rge and I also discussed the issue several months ago.
Terminology not withstanding, if we apply the "strict definition" of Intel's Thermal Specifications toward a more practical usage, and instead regard Analog Diode CPU temperature in the same context as Tcase IHS temperature, then we maintain the same 25c to 30c Distance to Tjunction Max, we all keep our chips chilled, and the n00bs don't assume that it's OK to run at 100c. If we look at one of the Scales in the Guide, it's clear how well the IHS Thermal Specifications in the Processor Spec Finder translates into sensible temperature scaling, as if Intel intended it to be as such:
Scale 2: Quad
Q9x50:
Tcase Max
71c, Stepping
E0, TDP 95W, Idle
16W
Q9x50:
Tcase Max
71c, Stepping
C1, TDP 95W, Idle
16W
Q9400:
Tcase Max
71c, Stepping
R0, TDP 95W, Idle
16W
Q9300:
Tcase Max
71c, Stepping
M1, TDP 95W, Idle
16W
Q8x00:
Tcase Max
71c, Stepping
R0, TDP 95W, Idle
16W
Q8200:
Tcase Max
71c, Stepping
M1, TDP 95W, Idle
16W
Q6x00:
Tcase Max
71c, Stepping
G0, TDP 95W, Idle
16W
-Tcase/Tjunction-
--70--/--75--75--75--75-- Hot
--65--/--70--70--70--70-- Warm
--60--/--65--65--65--65-- Safe
--25--/--30--30--30--30-- Cool
Nonetheless, I intentionally don't mention Tjunction Max values for any processor variants anywhere in the Guide, because my calibration procedure for SpeedFan is quite accurate without it, and based upon what we know at this juncture from Intel, Tjunction Max values are still very much in question for many 65nm variants. As always, my goal is to bring a better understanding of this crazy topic to everyone, so if there are any aspects of this discussion that I haven't considered, then please let me know.
Best Regards,
Comp