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

Socket Thermistors Debunked!

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

Colin

Arctic Silver Senior
Joined
Dec 19, 2000
Read this review. Obviously all socket thermistor based HS reviews are bogus. The other not so startling revelation is how poorly the Thermoengine performs. A slick piece of marketing it is, but it does this by fooling the socket thermistor and all of the inept reviewers that rely on the socket thermistor.
 

CalCoolage

Registered
Joined
Jun 30, 2001
Colin (Jul 11, 2001 11:34 p.m.):
Read this review. Obviously all socket thermstor based HS reviews are bogus. The other not so startling revelation is how poorly the Thermoengine performs. A slick piece of marketing it is, but it does this by fooling the socket thermistor and all of the inept reviewers that rely on the socket thermistor.

No comparison to the socket thermistor is given in the review, so the relation between that and the readings presented is unknown.
The Thermosonic performs closely to the other excellent HSFs (like the Alpha PAL6035), outstandingly good for an all-aluminum HSF, and it performs similarly, if not quite as good, as in tests that use the socket thermistor.

The review says:
"I recorded the ambient temperature around the heatsink using a second thermisitor while the testing was taking place. "

If this is what he did, then the temp difference depends on where the probe was positioned in the air flow from the HSF. Some heatsinks distribute the airflow differently than others, and the temp of the air coming off different sinks is different. In particular the air from heatsinks with a low air flow is hotter than others with a high airflow, so this positioning of the probe makes low air flow heatsinks look better than they would if actual ambient (room temperature) is used.

"To test the sinks I used a Dremel to drill a hole directly through the center of the heatsink where the core touches the bottom. If there was a clip in the way, I drilled through it too. "
Of the Thermosonic:
"The version used here is the one without the solid core, one of the very first versions. "
The unusual center of the Thermosonic perhaps performs some function. Drilling through it could affect the function, if he actually did that. If he did not drill through the center, then he measured the temp of a different part of the core.

"I then used a tiny dab of super glue to hold the thermocouple in place (the hole was so small it held itself in place for the most part). This made sure the tip of the thermocouple was touching the top of the core to get an accurate reading of the temperature."

In any case, drilling through a slightly different spot on different heatsinks measures the temperature of different parts of the core.

Drilling a hole in the heatsink creates a hot spot where the hole is, since that part is not in contact with the heatsink. The hole is small, but so are the .18 micron transistors, and heat conduction along the very thin cross-section of the chip is poor. Significant distribution of heat along the surface of the chip is provided only by the base of the heatsink. The expectation that top-of-the-chip measurements give the true temperature (whatever that is ) is not obvious to me.

I don't know that the Thermosonic fools the socket thermistor any more than, say, the Hedgehog or the Alpaha PAL6035. Although it might, I see no reason why it would. Blowing directly on the chip is a good way to cool it directly and does not qualify as "spoofing" in my book. One could equally well claim that blocking airflow around the chip, which most designs do, is a HSF design defect.

If I recall correctly, Anandtech or Tomshardware, quite a while back when the Thermosonic was new, did tests with a synthetic heat source and and on-the-heatsink thermistor that confirmed the rank of the Thermosonic as near the top at the time, pretty much debunking the idea that the in-socket thermistor is "spoofed".
 

Mac42

Member
Joined
Dec 18, 2000
So how does the thermister compare to actual temps when using water cooling? There wouldn't be any air-flow to mess it up... the only things I could think of that would chenge the temp readings would be a badly-cooled mobo chipset or very hot voltage regulators (both of which would heat up the pcb). I have my thermister bent up to contact the back of the CPU with Artic Silver bridging the two. I know it's still not 100%, but I don't feel like drilling into my Maze2 (and there's really no point since it will never overpower watercooling).
 
OP
C

Colin

Arctic Silver Senior
Joined
Dec 19, 2000
Perhaps I should have pointed these items out in my original post.

Drilling a hole in the HS and placing a thermocouple on the top of the CPU die is the method AMD uses when testing chips without an internal temperature sensor. This is the only AMD approved method and it is the method that was used in this article.

Quote from the article : “Thermal Integration ThermoEngine: With the stock fan this cooler doesn’t do so well, but with the Delta it is a different story. The price isn't bad on these and they just came out with colored anodized versions for those of you who are into that. Just be careful relying on the motherboard probe readings since they tend to be significantly lower (5-6C) than the actual temperature of your processor.” The reason for the temperature difference from the AMD approved method and the motherboard probe (socket thermistor) is because the open base design of the Thermoengine cools the motherboard and the motherboard probe. Here is more information on socket thermistors and the problems with measuring temperatures through secondary heat paths.
 
OP
C

Colin

Arctic Silver Senior
Joined
Dec 19, 2000
Lack of airflow on the motherboard due to watercooling will measure higher temps at the socket thermistor. They are still inaccurate because of the secondary heat path. My KT7A-Rs socket is completely insulated both front and back for use with pelts. With a straight Maze 2 block my temps read higher than on a noninsulated board because thermistor is some what isolated by the neoprene insulation.
 

CalCoolage

Registered
Joined
Jun 30, 2001
Colin (Jul 12, 2001 09:50 a.m.):
Drilling a hole in the HS and placing a thermocouple on the top of the CPU die is the method AMD uses when testing chips without an internal temperature sensor.
.
OK. I've debunked the AMD approved method. Removing the heatsink from part of the core elevates the temperature at that spot, invalidating the temperature reading.

>The reason for the temperature difference from the AMD approved method
> and the motherboard probe (socket thermistor) is because the open base
>design of the Thermoengine cools the motherboard and the motherboard probe.

If the probe is cooler, so is the chip it is in contact with, whether the mobo is being cooled or whatever the reason for the cooling. What is happening is the secondary heat paths are becoming more primary.

>Here is more information on socket
>thermistors and the problems with measuring temperatures through secondary
> heat paths
This article uses an electrical analogy, with heat flow being like current and temperature being like voltage. The current (heat) flows through different resistances. There are parallel current paths, two corresponding to the flow through the front and through the back of the chip. Each path is a series of resistances and there are various voltages at each resistance, referenced to ground (room temperature). The paths are like voltage dividers, so that the voltage measured (temperature) is a portion of the source voltage (temperature). The author calls the voltage division "compression".

If the heat flowing though the secondary path is insignificant, then measuring temperature in the secondary path gives a very accurate measure of temperature, just as when measuring voltage with a high impedance voltmeter, although the measurement is only proportional to the temp and not the actual temperature. OTOH if the heat flowing in the secondary path is significant, cooling the secondary path is an important way of cooling the CPU, and some HSFs provide additional valuable cooling. In that case, it also means the temperature readings are out of proportion. If some HSFs can cool the secondary heat path more than others, then they are also changing the thermal resistance in that path.

All temperature measuring devices, including an on-chip diode, at best measure something which is proportional to the temperature and not the actual temperature, so "compression". is not a phenomenon unique to the in-socket thermistor.

The link assumes there is such a thing as "the core temperature". But it also asserts that the temperature at the side of the case is different then the center, which is no doubt true. In other words there is no one temperature that is "the core temperature." Take for instance the on-chip memory cache. We know from the difference between the Athon and the Duron that it is a very large heat source and generates a lot more heat when it is accessed heavily than otherwise. It has a particular location on the chip and its temperature will be different than other parts of the chip.
An on-chip diode also has a particular location and can only possibly measure the local temperature, which is not necessarily proportional to the average temperature of the entire chip.
In short, the deficiency of the in-socket probe when compared to other possibilities is not so large, but rather the comparable uncertainties of alternative methods .have been neglected.
 
OP
C

Colin

Arctic Silver Senior
Joined
Dec 19, 2000
So AMD's emgineers don't know what they are doing? Joe's tests at Overclockers also point to the truth about socket thermistors. You can learn more of the truth here.
 

Kingslayer

Senior Member
Joined
Jun 12, 2001
Location
Port Charlotte, Florida
Of course their not accurate. Do you think you could pick up a quality motherboard for $100-$150 if the thermistor was spot on. Companies pay $1000's of dollars for temperature monitoring equiment. Places like refineries, and plastic companies. The thermalcouples that we used for testing ran about $200 a piece.

Would you really like to pay an extra $200 for your board just to have you temps 99% accurate instead of 95%?
 

Nevin

Mad Scientist from Arctic Silver
Joined
Feb 27, 2001
CalCoolage,

One thing that is important to remember about all temperature sensing devices, thermocouples, thermistors, and thermometers, is that they do not measure the temperature of what they are touching, they measure the temperature of themselves.

An in-socket thermistor has approximately 1% to 5% of its area in contact with the bottom of the CPU and 95% to 99% of its area in contact with the air in the socket. It is then attached via short electrically and thermally conductive wires to the motherboard.

You cannot accurately measure the temperature inside a wood burning stove by attaching a thermistor to the outside of the metal wall, you cannot accurately measure the temperature inside a house by taping a thermometer to the outside of a window, you cannot accurately measure the temperature of coffee by holding a thermistor against the outside of the cup, and you cannot accurately measure the internal core temperature of a CPU by touching a thermistor or thermocouple to the outside of the core.

This is not conjecture and supposition, it can be easily proved by physics, thermodynamics, math and actual comparisons of internal core vs. in-socket thermistor readings.

Nevin
 

CalCoolage

Registered
Joined
Jun 30, 2001
Colin (Jul 16, 2001 10:36 p.m.):
So AMD's engineers don't know what they are doing?

AMD does not make the claims of the infallibility of their procedure that people seem to assume they do. Really they make no claim whatever, but simply describe it. Spec sheets are designed for other engineers who deal with the the pitfalls, limitations and errors of testing every day. If the procedure were not specified, an engineer could very well do his tests differently, come up with different results, and have to determine through trial and error how AMD really did their test .

Ever compare the detailed results of the various websites and wonder why there are some inconsistencies? I've puzzled over some of these because the procedure looked pretty good while the results looked odd. One of the ideas is to use a synthetic heat source. What's the problem with that? Looking at one guy's setup, who was conscientious enough to show a picture, 75% of the heat source is not in contact with the heat sink, since the interface is through a plate the size of an Athlon slug. He used a heat spreader to make sure the heat could get to the interface, which was good thinking, but that provided a large extra area for dissipating heat. As we know, different HSFs inadvertently direct different amounts of air to what would be the chip carrier in real life, but in this setup that was really a second heat sink!

People sometimes put insulators over the artificial heat source. But spongy insulation compressed to 1/16 inch is a fairly decent conductor of heat, which would conduct heat around the fake slug and to the sink. Unlike electrical insulation, in order to do much, a heat insulator it has to be THICK. That's why you need 2 feet of insulation in your attic instead of 1/16 inch.

Since the synthetic setups don't use a socket, they make their own clip attachments. One guy reported a heat sink failure (where the temp rose without limit) like it was just a lousy HSF. From the picture, you could see that the clip would not press on the spot over the center of the fake CPU slug, since his imitation had the nubs symmetrical on either side of the fake die, unlike the offset real, live nubs. Since the clip did not press properly on the sink, the HSF failed. But it would work OK in real life.

One of the claims for a synthetic heat source is that they KNOW accurately the amount of heat delivered. Do they? How much of the heat really goes though the fake slug and not elsewhere?

Most often web sites do not give enough information to determine what might be wrong (or just different) with their procedure, if anything.

>Joe's tests at Overclockers also point to the truth about socket thermistors.
Yeah. I've been reading the new truth, which got me thinking about the new truth. How much better is it than the old truth? Once someone convinces me I have been wrong about something I have always accepted, I start to get real curious.

My apologies if any of this stuff sounds personal or offensive.