• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Intel C2D/Q/X Thermal Designs Explained

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

diaz

Member
Joined
Aug 27, 2007
Location
Canada
"What temperature is safe for my processor?"
This question has been asked once too often. I had made alot of research and found all the answers I needed. I still have some details to put out, explaining this more clearly.

Today I called two seperate Intel techs to get correlating answers on all this:

Thermal Monitor (TM): Function activated within the CPU once it has reached the maximum acceptable value at the Tj. At this point it will become active, and start a procedure called "PROCHOT#". It reduces the operating frequency of the CPU to help dissipate accumulated heat. It will remain in this state until it has reached operational temperatures again.

TM.2 (Enhanced TM): If TM fails to cool processor back down to operational temperatures, it will reduce voltages to VID reference voltage. Simultaneously, the CPU will be set to its lowest multiplier.

THERMTRIP#: If TM2 fails to cool CPU to reference Tj temperatures, all power will be cut from all cores. This will remain until temperature is once again restored under specified Tj.

VID: Factory set voltage used when CPU surpasses Tj. If voltages are in excess of VID, they will be lowered to VID in increments allowed by motherboard's BIOS.

Tc (Tcase - Intel labs): Temperature measured at the geometric center of the surface of the CPU. The only way to measure it is to crave a hairline groove in the CPU and apply/solder a hair-sized thermostat probe into the CPU's IHS. This is the way Intel measured the temperature to give the Thermal Specification of each CPU.

Tcase (On-die): This temperature is measured between the cores. It is a sensor included in every Core 2 Duo and Quads. This reading shows up as "CPU Temp" on most programs. It is a little hotter than the Tc from the Intel labs "Thermal Specification"

Tj: Is measured right in the CPU's die, the main silicon area.
The temperature sensors on each core of a Core 2 processor are not very accurate. They produce higher than actual readings to ensure safer operation of TM and TM2 - which in turn almost garantee longevity and reliability of Core 2 CPUs.
Quoted from Intel's processor finder webpages:
Thermal Design Power: (Also referred to as Thermal Guideline) The maximum amount of heat which a thermal solution must be able to dissipate from the processor so that the processor will operate under normal operating conditions.

Thermal Specification: The thermal specification shown is the maximum case temperature at the maximum Thermal Design Power (TDP) value for that processor. It is measured at the geometric center on the topside of the processor integrated heat spreader. For processors without integrated heat spreaders such as mobile processors, the thermal specification is referred to as the junction temperature (Tj). The maximum junction temperature is defined by an activation of the processor Intel® Thermal Monitor. The Intel Thermal Monitor’s automatic mode is used to indicate that the maximum TJ has been reached.
Thermal design power and Themal Specification are used in conjunction with each other as a reference to system builders. Thermal desing power is the amount of power drawn from the PSU, and a requirement for a motherboard. This means that the power consumtion indicated is only a worst-case scenario in which the proper parts are required to handle the power draw.
Thermal specification is NOT the maximum allowed temperature for safe operation. It is the temperature reached at the specified TDP, in the worst-case ambiant temperature (which is 38C for C2D and C2Q - 39C for C2X and C2QX), using stock HSF. Ambient temperature is measured at 2mm from the blades of the active stock HSF.


Sensor on Core 2 Quad processors:

There are 3 sensors per dual core. One on each die called 'DTS' or digital thermal sensor - which are more accurate than I have previously stated; which measure Tjunction (CPU Hot Spots) values. They are set at +/- 1C. But that is only avail on Kentsfield processors.

The Tcase value is measured at a sensor which is located right between the cores. The Tcase value is 15C under Tjunction value.

Tjunction values:

Tjunction Max = 85c (B2 Stepping)
Tjunction Max = 100c (G0, L2, M0, B3 Stepping)
More...
 
Last edited:
Lets look at some popular specs to keep misunderstanding to a minimum.

Q6600 G0 vs. Q6600 B3 for example:

G0 = 95W @ 71C
B3 = 105W @ 62.2C

Those figures are relative. That only means that during final testing of Intel reference designs, the B3 stepping of the Q6600 reached 62.2C when using full 105W of power. This also means that the G0 reached 71C at 95W of power. This obviously does not make sense from your point of view. But what you must understand, is at the time of testing, these silicon wafers are most likely from different batches. Also there are other factors such as relative humidity, and the fact that those measurements are at sea level, the flatness of CPU, the application of TIM, crappy heatsink, people testing, just to name a few.

As for the power consumption, looking at data sheets and general VID specs on G0 and B3 Q6600's, you will see that G0's have lower VID's on average. Lower voltage means lower power consumption. Essentially, G0 stepping is an under volted B3.

To finalize those specifications, they must submit the highest temperature reached by an average of the several reference designs. So really, getting a cool running CPU is mostly due to luck of the draw (silicon wafer), and also the wattage rating (TDP).

Lets look at a graph to clarify this:
QXTemp.gif

Here is a QX6700. See how the graph describes the power to temperature reference? This shows that when at 110Watts, the QX6700 will reach 61.1C based on this RCFH-4 reference from intel. And if you look carefully, when at 130watts, the temperature reaches about 64C. Familiar? That is the "Thermal Specification" given by intel as a reference, to what their prototypes were reaching on average. So these measurements are real-life ones made in Intel's lab by people like you and me.

Intel references are usually on par with production CPUs, but several changes can be made prior to prodution; which are done most of the time.

So what can be said about these thermal specifications and design power, is that they are a worst-case scenario of an example to base yourself on as a system builder/maintenance.

More...
 
Last edited:
What does all this mean for us?

Coretemp and multiple other programs, read the CPU's temperatures right at the cores. Those temperatures are Tjunction temperatures. The temperature of a single core is not very indicative of the tests performed in Intel's labs. Their measurements were made directly on top of the processor. So there is quite a difference between their measurements and the ones displayed in Coretemp.

The one intel based their measurements on are alot closer to "CPU TEMP" on most programs such as "PC PROBE" and others which display only 1 temperature no matter how many cores are present. *EDIT* This reading comes from a sensor on each CPU die, between the cores (this means that C2Q would have one on each chip). It is not as acurate as the Intel's Tcase values, it is a bit hotter. But usually within about 5C-10C from the Intel Tc, unlike the Tj which can be more than 15C off.

**EDIT - Note
C2D's feature 3 sensors. 2 of them are located directly in the hot spot of each core (Tjunction), and the last one is situated right at the center of both cores (Tcase). Not to be confused with Intel's Tcase measurements mentioned later with an actual groove in the center of the CPU where the measurements were taken. These measurements will be cooler since they are further from the CPU, but I will talk about that later.

This means C2Q has a least 5 sensors. 2 on each core for sure. I will try to find out if there is a sensor between each core, for a total of 6, or if there is only 1 sensor placed at the center of both dual cores.
**

This would be a good example (Q6600 B3):
Ambiant - 23
Idle - 44-44-40-40
Load - 58-58-55-55 (Hot spot of CPU - Tjunction)
CPU Idle - 38*
CPU Load - 52* (Internal CPU Tcase)
*PC Probe

So 38C and 52C in this example would be much more accurate when comparing to Thermal Specification from Intel. Since the provided reading is from the CPU die which is closer to the cores (right in-between) than the center edge of the Intel Tc is, the Intel Tc values would likely measure about 35C Idle and 48C load for example. This is much lower than the in-lab value of 62.2C specified by Intel. That is most likely because of an after market HSF used instead of the one provided by Intel.

More...
 

Attachments

  • kentsfield_chip.jpg
    kentsfield_chip.jpg
    16.9 KB · Views: 12,385
Last edited:
Example of an Intel Tc measurement preparation:

Q6600Tc.gif

Edit:
This measurement differs from the on-die CPU thermal sensor, which is located in-between each dual-core chip.
 
Last edited:
So the big question: How do I make sure my CPU is safe?

Looking at the Tjmax of a Q6600 for example (100C) - Intel suggests 5-10C buffer for temperature. Since temperatures vary from night to day, and season to season, it is recommended to make your measurements at the very hottest ambient temperature, and then make sure you are at least 10C away from 100C Tj.

So if you currently have C2Q:
Ambient: 25C
Load Tj: 75-73-68-68

Then calculate your peak ambient in your area, say 38C on a hot summer day. This would add (38-25=13) 13C over the hottest core.

75 + 13 = 88C

88C is under 90C which is already a safe mark. So from there on, you know your temperature should not be an issue, unless something happens to your HSF or TIM. But that's what TM and TM2 are for !

So in relative terms, for the same setup, PC PROBE for example would display something like this:

Ambient: 25C
Load CPU: 65C (approx)
Theoretical Tc: 62 to 64C

62+13 (safety delta) = 78C

EDIT: Intel measures @ 38C ambient @ 62.2C Thermal spec. The closest measurement to this would be using a program which displays an overall "CPU temp" instead of individual core temperatures. 62.2C thermal spec would probably result in individual core temps of about 72-77C. This is pretty close to our OC'd temps. 72-38 = 34C - Which would be a common delta for a stock intel cooler.

OVER VOLTAGE EFFECTS ON CPU

Over-Voltage is all about stressing the CPU's normal operation (VID) voltage. It only allows more power to be dissipated, to a certain extent. The amount of voltage you will be able to push beyond that is about 10-15%. That is basically the exact amount you would be able to push general diodes and transistors before they completely cannot operate/regulate properly anymore. Going higher than specified voltages for diodes is called "Breakdown Voltage". The level of damage cause to transistors by doing so might cause damage, but that is at owner's own risk.

As per low-heats with OVOC, lower heat will not save an over-volted chip. Over voltage, like I previously explained, will only be exceeding the transistors capacity and cause breakdown. Temperature will only affect this breakdown voltage by a small/negligible percentage. Extreme temperatures however, could cause the breakdown voltage to decrease considerably, making it easier to damage the processor.

Safe voltage is always stock voltage, but 5-10% OVOC is considered safe.
 
Last edited:
Very interesting. So, what is the conclusion? Can we run these chips hotter than previously thought? Should we be trying to keep what PC Probe (or equal) is reporting at or below (how far below) Intel's Thermal spec? Or would keeping CoreTemp's readings below (how far below) Intel's Tjunction spec be more accurate? Maybe we just need to make sure the CPU doesn't throttle?

And beyond all this what is the most voltage we can pump into these chips? Does it vary by stepping? Does super cooling really allow "safer" over-volting when the CPU will automatically reduce temps by throttling (speed and/or voltage) if it has to anyway?

These are the questions I want definitive answers to.

Edit...it looks like you've added something. Let me continue reading...

Edit2...temperature questions answered reasonably well so far. What about voltage?
 
Last edited:
How to find your CPU power consumption:

Pd = Power dissipation
C = Capacitance
V = Voltage squared
f = Frequency

Pd = C x V2 x f

First find stock capacitance @ VID:

C = Pd / V2 x f
C = 105W
/ (1.3125v(2) x 2.4)
C = 25.3968

Now add your volt and freq:

Pd =
25.3968 x 1.2V(2) x 2.8Ghz
Pd = 102W:eek:

Interesting... Very interesting...

Let's try a lower voltage (theoretic of course):

Pd = 25.3968 x 1.15V(2) x 2.8Ghz
Pd = 95W:drool:

These are theoretic, but gives you a guideline of what to expect !

On the other hand lets try very low voltage, stock clock altogether:

Pd = 25.3968 x 1.05V(2) x 2.4Ghz
Pd = 67.2W :soda:

So Bing's UVOC guide is an excellent way to decrease power consumption and improve temperatures ;)

Thats the same TDP as a stock E6300 ! Nice? You decide...

EDIT:

Conclusion:

Steps to a safe OC / Temp:

1. Look up your temperature using coretemp, everest, or any program that will give you a reading of each individual core.

2. Take the highest core reading @ load as the valid reading (ex: 75C).

3. Make sure to keep 15C under the maximum Tjunction for your CPU stepping.
ex: 100C is Tjmax for Q6600, so 100-15 = 85C

4. Factor in varying temperatures throughout the time you will be operating your computer at those settings.
ex: for a 24/7 rig, I would factor in the hottest day of the year - for a 2 day OC, I would factor in the possible rises in temperature from night to day

5. Calculate your results: 100C (Tjmax) - 15C (safe buffer) - 5C (expected temp rise) = 80C(maximum allowed full load temp)

6. To verify that your temps are safe, run programs that detect calculation errors - like super pi and prime, orthos etc.. If you CPU is 100% stable, then you have a safe temperature range.

7. Voltage: No matter what temperature you are operating at, excessive voltage could still fry your cores - which could happen if your PSU produces a fluctuation in voltage that would cause a dangerous bump. To be voltage safe, always make sure to back off from your max stable voltage. (Or get a good UPS - uninterruptible power supply)

8. Are your temps too high ? UVOC (under-volt OC) Less voltage means less heat. Try lowering your voltage to stock, or even under stock. Most under volted Quad cores can still clock over 3Ghz.

9. Detecting a bad cooler or HSF: UVOC your CPU at stock frequency, until your voltage is as low as it can reach. Some boards will allow 1.0V as a setting, some will stop at 1.1Vcore. At this point, look at your Idle temperatures, and full load temperatures. If the difference is more than 10C, you might have a bad cooler. Try re-seating the cooler with a good thermal solution like AS5 (asrctic silver 5), and try again.

-D
 
Last edited:
Very interesting. So, what is the conclusion? Can we run these chips hotter than previously thought? Should we be trying to keep what PC Probe (or equal) is reporting at or below (how far below) Intel's Thermal spec? Or would keeping CoreTemp's readings below (how far below) Intel's Tjunction spec be more accurate? Maybe we just need to make sure the CPU doesn't throttle?

And beyond all this what is the most voltage we can pump into these chips? Does it vary by stepping? Does super cooling really allow "safer" over-volting when the CPU will automatically reduce temps by throttling (speed and/or voltage) if it has to anyway?

These are the questions I want definitive answers to.

Edit...it looks like you've added something. Let me continue reading...

Edit2...temperature questions answered reasonably well so far. What about voltage?

Give him time to finish...I'm also keen to see the bottom line;)

*Edit* OK, looks like he's done...
 
Last edited:

Cooler is always better, but I say that 75C is a safe bet 24/7 for a Q6600.


-D

Ahh, this is what I've been waiting for...so if we keep the max core temp below 75 C using Coretemp for a quad, it would be safe to run 24/7...right?

I'm a simple man and need a simple answer:)

Bing...your comments pleeeze !
 
I already have an excel spreadsheet I made to calculate heat energy, maybe I'll upload it.
 
I guess I better re-enable TM in the BIOS!

I don't see a TM2 option in the BIOS on this board, but I seem to remember it on that DS3R board I had for a day...can anyone confirm?
 
Cooler is always better, but I say that 75C is a safe bet 24/7 for a Q6600.

It seems from your argument that the Xeons have the same safe temp then...~75*. The 85* vs. 71* Thermal Spec argument is meaningless now! I still like my chip, though. :beer:

Over-Voltage is all about stressing the CPU's normal operation (VID) voltage.The amount of voltage you will be able to push beyond that is about 10-15%.

Safe voltage is always stock voltage, but I would say 5-10% OVOC is considered safe.

I'm at 13.25% over VID. I guess the higher VID chips can take more voltage by this logic. Kind of makes sense. I've got a high VID, but my temps are not that high considering how much voltage I'm pushing and the fact I'm on air.
 
75C at the Cores with CoreTemp or 75C CPU Temp with any other monitoring program?

75C at the Cores for a "safe" temp seems kind of lower than I expected after reading all of what you typed up.
 
Motion for sticky seconded, this is exactly the sort of information that people need before the overclock. A non-electrical-engineer version of the voltage post would be nice though, something us mere mortals can understand.
 
Motion for sticky seconded, this is exactly the sort of information that people need before the overclock. A non-electrical-engineer version of the voltage post would be nice though, something us mere mortals can understand.

Agree, simplicity would be prefered...:)
 
Back