After posting our recent Prescott Overclocking article the sharp eyes of many review sites, online retailers, and enthusiasts noted some large temperature differences between our articles. If you have not been following our Prescott coverage let me give you a quick run down. The initial Prescott review was written by me (Nathan Kirsch) using the ABIT IC7-MAX 3 and the overclocking article was written by Jason Petermann on his ASUS P4C800-E Deluxe board. There was a large difference in temperatures noted between the two boards and due to the feedback from this I've spent the past couple weeks running our 3.2E Prescott on both boards. After using identical settings for testing we found that the results are not only surprising, but they also point out an issue that needs to be better understood by enthusiasts!
How "Core" temps are found:
Since this whole situation started with our Prescott the it will be the processor used for this example. Many people think that core temperatures found using Mother Board Monitor or other utilities is done by a thermal probe on the motherboard. This is possible, but with the boards in question today that is not the case. There is no temperature sensor found on the board in the socket area of the ABIT IC7-MAX3 and the ASUS P4C800-E Deluxe! The reading actually comes from a thermal diode found in the Prescott die.
We found that the Intel has incorporated two methods of monitoring die temperature in the Prescott; the Thermal Monitor (also known as: "Prochot/Thermtrip sensor") and the thermal diode. To better explain what these do it's best to quote directly from Intel:
What measures the temperatures:
"There are two independent thermal sensing devices in the Pentium 4 processor on 90 nm process. One is the on-die thermal diode and the other is in the temperature sensor used for the Thermal Monitor and for THERMTRIP#. The Thermal Monitor?s temperature sensor and the on-die thermal diode are independent and physically isolated devices with no defined correlation to one another. Circuit constraints and performance requirements prevent the Thermal Monitor?s temperature sensor and the on-die thermal diode from being located at the same place on the silicon. The temperature distribution across the die may result in significant temperature differences between the on-die thermal diode and the Thermal Monitor?s temperature sensor. This temperature variability across the die is highly dependent on the application being run. As a result, it is not possible to predict the activation of the thermal control circuit by monitoring the on-die thermal diode."
From the above statement it's clear that the Thermal Monitor is used to control the processor temperature by activating the TCC (Throttling) when the processor silicon has exceeded its maximum operating temperature. The TCC reduces processor power consumption as needed by modulating (starting and stopping) the internal processor core clocks. Other than the Thermal Monitor Intel placed a thermal diode on-die to give end users a way to monitor temperatures of the core.
How the temperature is figured:
"The processor temperature is determined through an analog thermal sensor circuit comprised of a temperature sensing diode, a factory calibrated reference current source, and a current comparator (see above image). A voltage applied across the diode induces a current flow that varies with temperature. By comparing this current with the reference current, the processor temperature can be determined."
The temperatures we see using MBM5, ASUS Probe, and other monitoring programs come from the on-die thermal diode. Thanks to a thermal sensor (ie: Super I/O chip) located on the motherboard the data from the thermal diode can be used to monitor the die temperature of the processor for thermal management. This thermal diode is separate from the Thermal Monitor?s thermal sensor and cannot be used to predict the behavior of the Thermal Monitor.
What influences the reading:
Every processor has different amounts of leakages and is therefore unique. Since each processor is unique the readings between two processors should never be the same. One of the parameters used in the equation to determine the CPU temperature is the Diode Ideality Factor (A.K.A. the Non-Ideality Factor) which describes the behavior of the diode relative to a theoretically perfect diode. The ideality factor depends on the characteristics of each individual processor and will vary slightly from one chip to the next. According to the data sheet the range of ideality for the Prescott ranges from 1.008 to 1.0137. The temperature calculation should take this range in to account in order to improve the accuracy of the reading.
One other variable that influences the temperature accuracy is the "'Series Resistance", which is a measure of the resistance in the traces leading up to and away from the thermal diode. The Prescott data sheet shows that this ranges between 3.242 and 3.594 ohms. Some diode sensors have the ability to adjust for the temperature error caused by the resistance, some do not. Lastly, cross-talk from high speed signals to the thermal diode traces might also introduce an error.
Now that you have the basic understanding where the "core" temperatures are coming from lets get testing!
- Intel Pentium 4 3.2E & 3.2C 800MHz FSB (Intel ES unlocked Samples)
- ABIT IC7-MAX3 & ASUS P4C800-E Deluxe Canterwood Motherboards
- nVidia 5900 Ultra 256mb (nVidia refference)
- Corsair XMS PC-4400 Memory
- Corsair HydroCool 200ex (Water Cooled) w/ Arctic Silver Ceramique TIM
- Seagate 120Gb Serial ATA 150 Hard drive
- Antec TruePower 430W Power Supply
- 17" Sony Trinitron monitor
- NVIDIA Detonator 53.03
- Intel Chipset Software Utility 5.00.1012
- ABIT BIOS Beta 14 / ASUS BIOS 1015
- ASUS Probe 2.22.04
- ABIT EQ 1.10W
- MBM 188.8.131.52
TEST 1: Overclocked Prescott at idle
For this test we ran our Prescott at 250MHz FSB w/ 1.5875Vcore giving us 4.0GHz.
ABIT Results: (Click image for full size)
ASUS Results: (Click image for full size)
Results: If you took a close look at the images you will notice that the ABIT board recorded idle temps at 50C while the ASUS board was showing 36C on both MBM5 and ASUS Probe. This is a 14C (25F) difference between the boards. Both boards have the Vcore set to 1.5875V in the BIOS, but are running the same actual voltage of 1.568V. Both the ASUS and ABIT boards record the motherboard temperatures around 26-28C. Using the same monitoring program, processor (Important due to the fact that each CPU has different amounts of leakages), voltage settings, and testing conditions we found a temperature difference to great to ignore or call "random error".
TEST 2: Default Northwood at idle
Since the Prescott is a new processor based on a new core the next logical step is to look at a known processor. For this test we ran our 3.2GHz Pentium 4 Northwood processor at defualt settings.
ABIT & ASUS Results: (Click images for full size)
ASUS "Default" temps
ABIT "Default" temps
Results: Again we see vast temperature differences between the CPU temperatures and this time the Motherboard temperatures are identical. The ABIT board came in at 36C while the ASUS showed 23C. The CPU temperature on the ASUS board is actually 3C lower than the board temps! The 13C difference here goes along with the 14C gap we noted using the Prescott. It should be noted that on this test we could not get the VCore to be identical, but even with the ASUS board having almost .1V more the temps were much lower than the ABIT's reading. MBM5 confirmed these results, but we put in the above pictures so we don't put you to sleep. We also used Intel's Canterwood "Bonanza" motherboard and found CPU temps to be around 27C. We found that these are three motherboards with three very different temperature readings.
This test eliminated the Prescott as causing inaccurate readings.
Putting It Together:
It turns out that the CPU or "Core" temperatures are not a measure of actual temperatures. The temperature is actually ?calculated? by the chip that is associated with the thermal diode. There is a parameter for calculation that is provided by the I/O chip vender, which is supposed to be standard, but that doesn't seem to be the case. It has also been determined and confirmed by Intel that the temperatures observed here have too much variation to have been calculated using the same parameter. This means that the issue lies in the inconsistency of the parameters written by different companies.
We also found that it is not possible to predict the activation of the thermal control circuit (TCC) by monitoring the on-die thermal diode via software utilities. This is because the Thermal Monitor?s temperature sensor and the on-die thermal diode are located at the different place on the silicon. The temperature distribution across the die can result in significant temperature differences between the on-die thermal diode and the Thermal Monitor?s temperature sensor. Making it virtually impossible to predict when the TCC will engage by monitoring the on-die thermal diode.
The only way to make sure that board temperatures are correct (at least close) and consistent is for Intel to better regulate a standard parameter that motherboard manufacturers and I/O venders must follow. Otherwise we will continue to see large temperature differences between brand lines. At the same time I also feel that it's not Intel's job to police how the venders implement their monitoring technology. This isn't a perfect world and perfection takes time and time is money, so the end user should get used to seeing core temperatures that are more than likely far from accurate.
So the next time your buddy is bragging that his cooling solution on an ASUS board is better than the one on your ABIT board just laugh and say whatever! If he doesn't believe you show him this article and put him in his place! At the current time it is not possible to consider comparing temperatures between two different brand lines.
Legit Bottom Line: The core temperatures that we "monitor" are inaccurate and are nothing more than a mathematical equation that varies due to the fact that board & I/O chip venders don't use the same equation!
- Intel Pentium 4 Processor on 90 nm Process Datasheet
- Intel Pentium 4 Processor on 90 nm Process Thermal and Mechanical Design Guidelines
A Big Thanks To The Following People/Companies For Letting Me Bug Them During My Research:
- Intel Public Relations
- Intel's Thermal Engineering Dept.
- ABIT USA/Taiwan
- ASUS USA
- Joel @ Sudhian