SUMMARY: The correlation between in-socket thermistors and CPU case temps is tenuous at best and misleading at worst.
I routinely get emails from readers saying something like “But my Motherboard Monitor temps are different from your heatsink test results.” We, and others, have done a fair amount of testing on this very issue, but as one of our readers, Ariel Vera, pointed out:
“All your individual articles are *excellent* but they are usually published too far apart timewise for the reader to see how all these things interconnect and effect each in that delicate “ecosystem” we call our computer case.”
Excellent point! So to answer this vexing question, I decided to go back and pull together relevant data from a number of articles authored by us and other contributors.
Almost all users do not have sophisticated digital thermometers and thermocouples to measure CPU temps. What we do have, for current Socket A CPUs, is the in-socket thermistor. While motherboard manufacturers probably spent all of $0.05 to install it, many readers expect to see valid comparative results rivaling a $300 digital thermometer. This is not a valid assumption.
First, let’s better understand what is being measured where. This diagram shows the relationship between the in-socket thermistor to CPU temps we measure with thermocouples:
Ambient temps are measured one inch from the fan’s intake. The variance between temps here and MBM system temps measured some distance from the fan’s intake is enough to throw results off significantly.
The Heatsink Thermocouple is placed above the center of the CPU’s top. The CPU top is where all the cooling is taking place. Intel and AMD heatsink specs are geared to temps measured at the CPU’s case top.
CPU case temps are very closely approximated by drilling a small hole into the heatsink and inserting a thermocouple into it. By locating the thermocouple over the CPU’s center, the difference between the CPU’s case temp and the heatsink’s temp is due to the thermal coefficient of the grease or pad used between the two. Arctic Silver specs their thermal resistance at 0.0028, which means at 100 watts, the CPU case top will be 0.28C higher than temps measured at the heatsink.
So, for all practical purposes, temps measured at the heatsink closely approximate CPU case top temps.
Going down one step from the case top are internal core temps as measured by an on-die diode. Right now, Intel has this feature for PIIIs and AMD is implementing it in the Palomino. However, note that this silicon thermostat is INSIDE the core and is very small; Intel states that the temp variance between this diode and the hottest possible spot is 4.3C. In addition, because it’s buried in the silicon, I’ve seen something like a 4-6C difference from the diode temp to the case top.
Going down further from the CPU case top is the in-socket thermistor. Now, not only is this physically the longest distance from the CPU case top, it is, in some instances, not even physically connected to the CPU’s back. This thermistor is what is shown as “CPU temps” in BIOS and is what CPU monitoring programs like Motherboard Monitor use for all Socket A temps.
Nevin (Arctic Silver) stated in his article Why Many Thermal Measurements Are Not Valid:
“Many people, including many review sites, attempt to measure and quantify a Thermal compound change or heatsink change by measuring the side of the CPU case or the back of the CPU case. Both these measurement points are in secondary heat paths and are between components that present unknown thermal resistances. While measurements at these points may change some due to changes in the primary heat path, the degree of change will not necessarily be proportional to the actual amount of change in the primary path.”
We’ve done a fair amount on this issue – extracted below are results (details HERE) comparing a number of different heatsinks’ MBM and CPU case top temps:
- ABIT KT7, no raid
- CPU Stress: Prime95 – avg 15 minutes
- Fans: YS Tech 26 cfm; Super Orb 2 fans; Vantec 60x15mm
- Heatsinks drilled with thermocouple by Andy Lemont
- Temps measured by Omega HH23 Digital Thermometer
- Arctic Silver thermal grease
Temps were monitored for stability; each time temps were recorded, no substantial fluctuations were observed. In addition, at boot up, thermocouple temps were about at ambient, so there was no “residual” heat observed from earlier testing. The socket interior was insulated with closed cell foam which served to hold the thermocouple against the back of the CPU.
The following table lays out our test data:
All temps measured in Centigrade by Thermocouples except MBM417; Heatsink = probe drilled into heatsink; CPU was AMD T-Bird 1000 MHz at 1.82 volts – 57.4 watts.
Not much of a correlation, is there?
In an additional test, a comparison was made among another sample of heatsinks comparing CPU Case Temps, MBM and CPU Back Temps as measured by a thermocouple glued onto the back center of the CPU, with the following results:
While you would expect some close tracking between MBM and CPU Back Temps, this graph shows the lack of any correlation for a number of heatsinks on the same motherboard between CPU Back Temps and CPU Case Temps (and by proxy, MBM Temps):
But maybe this is due to some imprecision in measuring temps with thermocouples? OK – let’s see what Nevin (Arctic Silver) found (details HERE) using Intel’s on-die diode against MBM temps:
Same CPU, same motherboard, different heatsinks, and Nevin finds significant variation not only between heatsinks, but also finds that the MBM results are “compressed” compared to what is actually happening inside the core, by about a 3:1 ratio. When you see heatsink roundups and something like 12 or 15 heatsinks fall within 3C of each other, this is why.
This compression effect was amply demonstrated by Mike Warrior’s well researched article:
“AMD states that thermal resistance between Ccore to Cback (core to back of core) is 0.5. So on a KT7, where the drop from directly behind the core to core edge (where the mobo measures temp) is around 25%, a KT7 registers less than half of the CORE temp change.”
“Having examined the secondary pathway temp loss problem (not in detail, though), backside thermistors show only a fraction (less than half, probably something more like 35-30% on a KT7) of what the actual CPU core temp change is, which in turn leads to both Inaccurate Results and Inaccurate, Non-Relative Comparisons.”
The essential problem in trying to reconcile in-socket to CPU case temps measured at the heatsink is that each heatsink has a different temperature gradient between the CPU and socket thermistor, as shown above. The same motherboard will show different temperature gradients among various heatsinks. In short – apples to apples comparisons among heatsinks using the in-socket thermistor is misleading.
Considering the evidence, one approach to approximate this relationship is to use a range of values to reconcile temp measures between the two. Based on tests to date, a range of 4-6C for On-Die Diode Temps and 5-15 C for CPU Case Top Temps will contain most of the variations, as shown in this hypothetical example below:.
|HS C/W x watts|
|On-Die Temp Range|
|In-Socket Temp Range|
Considering the variability in observed readings, it is entirely conceivable that a heatsink with a higher C/W could show a lower MBM temp than one with a lower C/W, within some range of C/Ws. Further, in such a situation, a user may find that the heatsink with the higher MBM temp in fact allows higher overclocking (due to lower CPU Case Temps) than the supposedly “better” heatsink with lower MBM temps.
The in-socket thermistor is not designed to be a precision CPU Temp measuring tool; it is there to prevent your CPU from self-immolation by acting as an early warning tool. The further this tool is from the CPU Case Top, the more likely it will be affected by factors which render it useless for detailed heatsink comparisons on the same motherboard, leading to potentially erroneous comparisons.
Lacking precision measuring instruments, users might identify top performing heatsinks by their ability to overclock higher or run at lower voltages compared to another model.