An Interview with Arctic Silver Or How Most Things I Knew About Cooling Were Wrong.

[MrThompson]

This is an excellent read.

 

[MoFunk]

Nice read. Thanks.

 

[damocles]

Excellent read

 

[Mikewarrior2]

Gee, i didn't already know this


Mike

 

[Egrimm]

Good read but nothing new though.

 

[Wind]

Thanks. Nice to know.

 

[WarCon]

Good read, but I will have to give credit to Mikewarrior2 for helping me understand this when I first joined the forum. I was a little hard headed, but I get it now............

Temperature readings just plain don't mean much.................. Stability is the key.

 

[Insidious]

<< Temperature readings just plain don't mean much.................. Stability is the key.
>>



I almost agree with ya here. Where external temperature measurements are VERY valuable are when used as a comparison, within a given system to evaluate the effects of cooling design changes. Whether or not temp rises or falls helps you know if you are making progress or not.

On a side note, the same is true for the 3DMark2001 comparisons everyone wants to live and die for. Those marks are VERY helpful (WITHIN A GIVEN SYSTEM for evaluating the effects of changes made, but pretty much useless for comparisons of one machine to another.

So I agree, looking for stability is MUCH more valuable than looking for a certain number on your thermister output. After you find that stable point, down is better, up is worse.... that's about all ya got

 

[gregor7777]

Good read. Put a few ideas into layman's terms so guys like me can have an easy read.

 

[Mikewarrior2]

Insidious,

almost agree with ya here. Where external temperature measurements are VERY valuable are when used as a comparison, within a given system to evaluate the effects of cooling design changes. Whether or not temp rises or falls helps you know if you are making progress or not.


I have to disagree with that statement. AN external thermistor (in this case a socket-thermistor) is not a good enough tool for comparisons. Between compression of temp changes and the ability to effect only socket-thermistor readings while not actually cooling the CPU are definately present. When it comes to CPU temps, the best tool will always be the internal diode, as it isn't subject to heatsinks that cool the thermistor and it will show full temperature changes.

The same applies to comparisons wehter they are a "review" or even a single user. Again, because there are heatsinks that can drastically reduce socket-thermistor "readings" while not actually doing the same effect to the CPU core temps.


Mike

 

[Insidious]

Actually, I think you do agree.

I say you cannot compare outside a given system.

However, if I put on a different sink and look to see if my external (backside) temp goes up or down as a result, I have the knowledge of whether this sink change was a good on or not. (just dont know how much better or worse)


You need to knock the dust off your physics books if you think I can change my CPU internal temperature and cause a temperature change in the opposite direction on the backside temp sensor.

 

[Mikewarrior2]

You see, socket-thermistors are not isolated measurements. External factors effect them.

See this charts comparing socket-thermistor to die temp:

HardOCP versus OVerclockers.com Tests

And this chart comparing heatsink-thermocouple and socket-thermistor(both from Overclockers.com):

Heatsink Thermocouple versus Socket-thermistor

As you can see, backside measurements DO not always reflect changes in CORE temp or Heatsink-Junction temp changes. Also notice the lack of consistency between heatsink-junction readings and socket-thermistors. This "inconsistency" can certainly show certain poorer performing heatsinks to end up looking better when tested with socket-thermistor.


Mike

This is about phsyics books. I know the physics behind it. The fact is that you're talking about a good insulation layer, and in 90% of hte mb's, also an air gap between the CPU Core and socket-thermistor.

 

[johncar]

http://www.beachlink.com/candjac/index.htm TVanalogue.htm article proves Nevin's and Mike's case re socket thermistor/any external temp sensor compression with analogous DC electrical circuits using typical numbers/statistics. Give it a read.
John C.

 

[MrThompson]

Thanks John.

Colin

 

[Insidious]

You might want to be sure you understand test conditions and controls.

as much as we might like to think so.... Apples just won't fall up :/

(I'm guessing there are inconsistancies with airflow in the first graph. (induction vs. eduction, etc))

 

[johncar]

Insidious,
Just to set our record straight...the analog model does not show that the socket thermistor "reverses the polarity" of the internal temp change. It only confirms the so called "temp compression"...the lesser magnitude of change shown by socket thermistors.

However, note that the "heatsink" temp change does so in the opposite direction...albeit to a very small degree...caused by the change in percentages of flow in the different heat flow paths, the heatsink flow path gets more flow with more heat conductive interface materials, resulting in a slightly higher temp drop from hs to constant case air temp.
John C.

 

[Nevin]

Insidious,

With the in-socket thermistor, you actually can get the measurement to move one way while the actual CPU core temperature moves the other way. The in-socket thermistor does not report the temperature of the CPU back, it reports the temperature of itself. Its own temperature is affected by the CPU back, the air temperature in the socket and the temperature of the motherboard just behind it, where it is mounted. The thermistor's temperature (and the temperature it reports to the measuring chip) is an average of all these temperatures.

So it is very possible to switch to a heatsink that blows air down around the socket and cools the motherboard and the hot components around the socket and have the in-socket thermistor report a lower temperature while the CPU core temperature actually goes up. The opposite of this is actually fairly common with watercooling. Since the waterblock does nothing to cool the in-socket air or the motherboard, the CPU core can be several degrees cooler than with a standard HSF, but the in-socket thermistor will report higher temperatures.
This can be seen in the following test of several heatsinks and a watercooling setup.
Test at overclockers.com

A heatsink engineer recently told me that his company's new heatsink had been specifically designed to blow more air down on the motherboard around the socket than their old designs. He said that while the new model was actually only 1C better on a 60 watt processor, the in-socket thermistor would report temperatures 5C lower due to the new airflow. He joked that since almost all the hardware sites test with in-socket thermistors, maybe he should switch to a slower and quieter fan and sell the unit as being quieter than the model it replaced and 2C or 3C better. Of course with the lower airflow fan, the performance would actually be worse than the old model, but the in-socket thermistor would report lower temperatures.

I spent a lot of time measuring temperatures with various Intel PIII and Celeron II processors on a slotket that had been modified to include an in-socket thermistor as well as a calibrated reading of the CPU's internal temperature diode. Several heatsinks I tried gave lower temperatures on the in-socket thermistor while the actual CPU core temperature went up.

Bottom line...
The in-socket thermistor is provides a highly compressed temperature reading that is an average of the CPU back temperature, the socket air temperature and the motherboard temperature. This measurement is useful for determining if the heatsink is getting dirty, a fan has a problem, or a general vague idea of overall system cooling. This measurement should never be relied upon to determine the actual performance or full potential of a particular component of the cooling solution.

Nevin House
Arctic Silver, Inc.

 

[Mikewarrior2]

Insidiious,

That is precisely why no website should use socket-thermistor's for testing purposes. If something as simple as airflow coming off a heatsink can drop socket-thermistor readings several degrees C and not have the same effect on the CPU, then they aren't going to be accurate. Clever heatsink design can easily "fudge" results.

Pay close attention to Nevin's Link. The important number to note is the MBM temp, which is hte onboard socket-thermistor. If someone were to review this with a socket-thermistor, then WaterCooling would be 2C hotter in socket-thermistor temps, while 8.6C cooler in CPu-Heatsink Junction temps. You're seriously going to tell me again that it isn't possible for socket-thermistor to rise while core drops again? I've done plenty of "research" into this as well as trying to soak up as much info from various people who are highly knowledgeable in this field.

This isn't an isolated case. The graphs I provided further show examples of how a heatsink can be worse in core, yet show up better in socket-thermistor. IN an ideal world, one would hope that you could get an Analogous graph to actual Die or actual Heatsink-Junction temps. Unfortunately, you get a certain graph for the "real" readings (DIe and Hetasink-Junction), while socket-thermistor readings are all over the place and do not match or even follow remotely the core temp change over ambient.


Mike

 

[Insidious]

Very good information! Thanks to all.

 

[Duallys]

<< Insidious,
When it comes to CPU temps, the best tool will always be the internal diode, as it isn't subject to heatsinks that cool the thermistor and it will show full temperature changes.
>>


So ... how does one go about reading temps from the internal diode?

 

[Mikewarrior2]

Duallys,

Unfortunately, with AMD chips, you're gonna have to wait for mainstream diode reading support... A7V266E diode support is broken. Tyan Dual boards have shoddy windows support for diode reading. And no sign of a program from tyan to fix this. The MSI dual board has socket-thermistor . Hopefully by Thorougbred time internal diode support will work properly on MBs.



Mike

 

[Duallys]

Perhaps I should have been a bit more specific ... how you read the temps from the P!!! internal diode? It would appear that, on my P!!! MBs, the BIOS and/or MBM is picking up the socket thermistor temps ... did I miss a setting in MBM?

 

[Mikewarrior2]

Which p3 mb(s) do you have?



Mike

 

[Duallys]

<< Which p3 mb(s) do you have? >>


P3 Socket 370 MB(s) ... Tyan Tiger 230 and Gigabyte GA-6OXET.

 

[Mikewarrior2]

Duallys,

The tyan is a socket-thermistor based board... no internla diode readings available...

The gigabyte, on the other hand, should be able to read the internal diode.