It is possible to use Intel’s Thermal Diode found on its chips to estimate heat sink efficiency. Heat sink efficiency, measured as Degrees in Centigrade / Watts (C/W), in this case is the temperature drop measured from the thermal diode in the CPU to ambient air temp. The thermal resistance between the CPU core and the ambient temp is comprised of the resistance of the CPU core to its case top, the thermal grease between the case top and the heat sink base, and the heat sink itself.
We are not attempting to measure each component as some of them are difficult if not impossible to measure accurately. What is measured it the total efficiency of the cooling solution used. You need a motherboard which features thermal diode measurement. I used Motherboard Monitor 4.10 to retrieve temp data from the diode with a SOYO 6BA+IV.
To determine how closely thermal diode temp is to actual temp, I used a simple test; I ran the CPU with Waterfall while measuring the ambient temp at the heat sink fan’s intake. I found that the ambient temp and the reported temp by the diode were the same. When running Waterfall, 99% of the processor is shut down, so it should be basically a hunk of inactive silicon. As such, it should be at ambient temp, which in fact for the Soyo 6BA +IV it was. If not, include the difference in the equation below as a correction factor.
The Formula
Based on Intel’s methodology to estimate Thermal Junction to Ambient Thermal Resistance (p44 “PIII Processor Thermal Design Guidelines”), the following formula was used:
Heat Sink Efficiency (C/W) = (Thermal Diode Temp + Diode Worst Case Factor + Motherboard Correction Factor – Fan Intake Temp) / CPU watts
The Diode Worst Case Factor is “The worst case difference between the thermal reading from the on-die thermal diode and the hottest location on the processor’s core.” Intel uses 4.8 C for the PIII. I am using this for the Celeron also since the cores are similar and I have nothing better. I am also taking all readings using Prime 95 to stress the CPU as much as possible. Intel uses a CPU stress program which gets the CPU to 80% utilization – if you want to be conservative, reduce estimated CPU watts by 20%; naturally, the estimated C/W will be higher (not as good).
C 366 @ 366 MHz
Fan Intake Temp = 20.9 C
Processor Watts = 21.7
Thermal Diode Temp = 31 C (Prime 95)
Thermal Diode Correction Factor = 4.8 C
Heat Sink C/W = (31 + 4.8 – 20.9) / 21.7 = 0.69
C366 @ 500 MHz
Fan Intake Temp = 21.1 C
Processor Watts = 21.7 * (500 MHz / 366 MHz) = 29.6
Thermal Diode Temp = 36 C (Prime 95)
Thermal Diode Correction Factor = 4.8 C
Heat Sink C/W = (36 + 4.8 – 21.1) / 29.6 = 0.67
The heat sink used was the Global Win FAB 28. The difference between the two measurements is negligible; in fact, I am pleased to see that they are so close considering that the reported diode temps are in whole numbers. Regardless of CPU speed, cooling efficiency will be the same at any setting.
This methodology will allow users and testers to better estimate how good one heatsink really is compared to another. I hope it will also force heatsink manufacturers to publish “realistic” C/W measures since independent verification is available to many testers.
Many thanks to John Carcich for his insightful emails which helped to focus my thinking on this issue.
Be the first to comment