Now that AMD & [upcoming] Intel CPU's are using upwards of 50 Watts; it is vital for users to have a better idea of the true maximum temperature these silicon furnaces are running at. 

Burningissues has shown in a linked article how the default readouts from motherboard sensors are appallingly inaccurate. This is not the fault of the utilities - they just report faithfully the data they are fed.

Here we show how motherboard monitoring utilities - like the excellent [freeware] MBM5.x & MBProbe - can be calibrated without expensive widgets to show the maximum temperature your CPU runs at. 

The data-table applies to several common fansinks used on FC-PGA/Duron/TBird CPU's & contains only data we believe to be reliable: most of this present data originates from the generously shared work of Millennium Thermal Solutions - we will publish other reliable data based on real-world testing as it becomes available. 

You must be using one of the listed fansinks on one of these CPU's to be able to calibrate using this technique. 

Once calibrations have been set you can of course swap components - but temperatures ranging more than a few degrees ºC away from the temperature at which you calibrated will again be subject to the inbuilt errors of your motherboard's onboard monitoring setup.

Tools: 

System with a FC-PGA/Duron/TBird & one of the fansinks in the table below. 

Household thermometer [digital readout type w/ remote sensor best]

Table of Processor Specifications

A monitoring utility [MBM or MBProbe recommended]

Prime95 [there are alternatives: whatever you choose must load the CPU 100%]

Table below

Calculator

Formulae below

[all measurements/calculations will be in metric units; if you prefer to use your monitoring utility in degrees ºF set it to ºC whilst carrying out the calibration]

The first thing to do is to open your computer's case & put the household thermometer [or its sensor] about 50-75mm from the fansink's fan. It is best if the sensor's at the same height above the floor as the fan.

Start up your system & open your monitoring utility.

When you open "settings" then click "temperatures" & "alarm" of the current version of MBM you'll see something like this:

 - there's a similar "compensation" [+ & - ] in all competent monitoring utilities; if your present one doesn't have this, bin it & d/l MBM or MBProbe.

What you want to do is to adjust this compensation so the CPU temperature readout is set as accurate as possible after, say, 20 minutes' running  at 100% CPU load.

It doesn't matter what your local temperature is: we can find that out with the household thermometer. Your CPU @ 100% load will run at a fixed temperature above this local temperature - so long as we know the real-world tested efficiency of your fansink [its "ºC/W"] & how much power your CPU uses at maximum load.

Go to the Processor Specification's page & find your CPU [we'll deal with what to do if it's overclocked in a moment]: if an Intel FC-PGA, note down the maximum amps & the default vcore; if an AMD CPU note down the maximum watts & the default vcore.

Check the real vcore actually being supplied to your CPU - this will be in the "voltages" section of the monitoring utility & will likely be in the range 1.60-1.95v

Calculate the full-load Wattage of your CPU as follows:

[Intel FC-PGA - not overclocked] 

= max amps x real vcore

[AMD Duron/TBird - not overclocked] 

= max watts x [real vcore/default vcore]²

[Intel FC-PGA - overclocked] 

= [max amps x default vcore] x ([overclocked MHz/default MHz] x [real vcore/default vcore]²)

[AMD Duron/TBird - overclocked] 

= max watts x ([overclocked MHz/default MHz] x [real vcore/default vcore]²)

 - note your result down - call it: "Watts max"

Go to the table below & read off the ºC/W j-a for your fansink & h/s grease combo.

Multiply: Watts max x ºC/W j-a - make a note of the figure: call it: "Delta ºC"

Now run Prime95 in "torture-test" mode for at least 20 minutes - with this still running take a note of the thermometer-reading: call it: "Ambient ºC" 

Still with P95 running open up the "dashboard" of MBM - or the temperature display of other monitoring utilities - & make a careful note of the CPU temperature shown.

Add: Ambient ºC + Delta ºC - this is the calculated true temperature of your CPU at 100% load: call it: "CPU max ºC"

Reset the "compensation" in your monitoring utility until the displayed CPU temperature = CPU max ºC

 - that's it. Now you can stop P95. 

If you run your machine with the case open you can set the "system" or "motherboard" readout to = Ambient ºC; or if you run with the case closed you can try whatever means seem sane to set this. 

If you're lucky enough to have a remote probe on your mobo, lead it to within 50-75mm of the fansink's fan [put a small shield on the probe if possible] & set it to Ambient ºC - you can call this: "Air-Intake" in your monitoring utility.

You will likely see - because of the non-linearity of the way mobo-sensors work - that your temperatures at idle read a bit high - who gives . . . . . now your monitoring utility is calibrated you should at last be able to get tolerably reliable comparisons [at 100% CPU load] between CPU's - default & overclocked, fansinks, & h/s greases.

PLEASE TREAT THE TABLE BELOW AS PROVISIONAL & FOR INTEREST ONLY: WE DO NOT BELIEVE TCASE MEASUREMENTS REFLECT REAL-WORLD EXPERIENCE - THOUGH THIS IS THE "APPROVED" INTEL/AMD TECHNIQUE : WE ARE WORKING WITH OTHERS TO PROVIDE A FULL TABLE OF BCASE C/W MEASUREMENTS - PLEASE ACCEPT OUR APOLOGIES FOR ANY UNINTENTIONAL CONFUSION CAUSED.

Heatsink grease, its quality & thickness - which depends on clamping-pressure & on the surface-finish of the CPU die & fansink base - makes up a surprisingly large component of the total ºC/W j-a of the installed fansink: this is tough to calculate with precision. 

Above is a graph for FC-PGA/Duron/TBird-sized CPU's showing the influence this can have. If you spend money on a high-performance fansink of, say, 0.40 or better ºC/W s-a it seems silly to use a poor-performing grease which may, under the worst circumstances, double this total figure. The very high pressures being exerted by the latest clips may improve the performance of mediocre-quality compounds.

Careful polishing of the fansink base can improve overall performance, by allowing the thinnest possible layer to be applied - ideally with something like a stiff-backed razor-blade as the spreader. 

The huge advantage of the low-resistance greases is their flat performance, relatively uninfluenced by thickness or pressure. Any h/s grease with a W/m² figure of 3 or better should give fine performance; all those above 5 are effectively identical on polished surfaces. It is worth noting that copper/copper-insert fansinks may be corroded if a non-copper metallic grease is used long-term.

< Hot Air [analysis of measuring techniques] 

<< GOrbal-luvvin' [analysis of fansink web-tests]

<<< Taking your Computer's Temperature [howto]

copyright © shoarthing November 2000 for Burningissues.net - all rights reserved - with thanks to Andrew Lemont for argument & facts, Jim Fager for starting it all, & Joe Citarella for keeping an open mind.