Woohoo! Anand's got new heatsink testing methodology!!

[Quad]

Linky

seems pretty decent and accurate. what do you think (*wink wink* mikewarrior2 )

 

[cdternie]

Looks like a MUCH better, well-rounded way of evaluating heatsinks!

Another great job by anandtech =]

 

[lifeguard1999]

Great article IMHO, but it lacked in one thing. Results. ARGH! I was hooked reading the article, and eagerly waiting for the heatsinks tested and results. *sigh* I guess I will have to wait.

THG also has a heatsink article up today.

24 CPU coolers in review

While I disagree with the testing methodology, I do like the new kids on the block and their noise ratings.

 

[Quad]

yep
new hsf roundup tommorrow
/me can't wait!

 

[GL]

<< While I disagree with the testing methodology, I do like the new kids on the block and their noise ratings >>



I believe THG, AnandTech and a handful of other sites have agreed upon the same testing methodology.

-GL

 

[Mikewarrior2]

I dont' see THG in this list of people doing this project. I know Stephen, JOe and John, somewhat know Andy, and don't know TIllman out of the people building this one. Other than the fact that they're only at 43W(which is low, IMO), it should give pretty good results.

Secondly, I highly disagree with Tillman's comment that the P3/Internal Diode is not accurate enough for testing purposes. It is perfectly accurate for comparing heatsinks. And on top of that, the diode can be calibrated to provide actual core temperatures, rather than just showing the full &quot;delta&quot; between heatsinks.

I also disagree with all the &quot;good job Anandtech&quot; comments. First, it isn't an Anandtech Endeavor. They still have that old review in their database. How many new users have been influenced by that very, very poorly done article. IT was only after serious arm-wrangling in this very forum that some of us forced tillman's hand in doing a new change. Secondly, 4 out of the 5 people who did it are non-anandtech.But that was only part of the goal, the other being a revision made to the now 5 month old review. Thirdly, if issue was not made of the last review, Tillmann and Anandtech would hop along happily going with the other testing method.

I'm happy they made a change, but I'm not happy they left the old review unrevised because I still see plenty fo people following the last &quot;guide&quot; as the end-all be all of heatsink tests.


Mike

 

[marooned]

Well said Mike, now if only somebody would do some severe arm mangling/wrangling and get either abit/via to fix their ____ing AGP. I'm not the only one having problems here! *sigh*

 

[adinov]

I have read the new testing methodology, and it seems to me like someone hasn't done their homework!!

I don't mean to be critical, but any decent text on temperature measurement would reveal that thermocouples are just not well suited to measuring temperatures in this range. With their low sensitivity, they are best suited to measuring temperatures above 1000 degrees C.

The best sensors for this temperature range (-50 to 150 degrees C) are, believe it or not, semiconductors!!
Diodes are excellent when calibrated correctly.
Transducers like the Analog devices AD590 and later derivatives, and the NSC LM135 series and its relatives are the industry standard for accuracy (even uncalibrated accuracy), and reliability in this temperature range. The &quot;F&quot; package from AD590 is similar in size to many thermocouples.



If someone asked me to solve this problem, I would have done this:
Use a dud (or working) CPU as both the source of the heat, and as the measuring &quot;diode&quot;. Reversing the CPU's power supply polarity results in a diode that would spread heat very evenly all over the silicon die. This &quot;diode&quot; consists of all the transistors in the CPU reverse biased and it could also serve to measure the temperature!!!!!

I would use pulses of constant energy, that way, power dissipation is directly proportional to the pulse rate. The temperature can be measured by passing a small test current between the pulses to measure the diode voltage drop which is inversely proportional to temperature.

The &quot;diode&quot; would need to be calibrated with an unpowered CPU at the chosen test current. Linearity can also be measured to validate this methodology. For this, measurement at 3 temperature points is required. I would place the CPU in a watertight plastic bag and immerse in water with a pre-calibrated probe. Measurement should be performed at 0C, 100C and somewhere in between.


hope this helps!

 

[johncar]

adinov,
Would agree with everything you say except the knock on thermocouples.
We used type E thermocouples for power plant process temps from 70 to 1050F, and their individual calibrations were always minimal at the lower temps...(simply because as you know, error is a function of wire impurity and temp). TCs do have advantage of small size, (of bare wires bead), not interfering with cooling process...we use as small as .005&quot; wires TCs.

Problem with TC low temp is the need for precision readout equipment, (which you refer to), but that's only a matter of $$$$$. We use lab precision Fluke multimeter with ~.03% DC accuracy and 1 microvolt resolution...meter zero check specs at +/- 3 uv, meter actually reads +/- 1uv. Think Tillman is using an ~$220 Omega temp meter, so his error may not be significant.
John C.