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

[Duallys]

<< 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. >>



The Gigabyte GA-6OXET has a socket thermistor ... but, it should be able to read the internal diode?

 

[Mikewarrior2]

The GA-6OXET has a socket-thermistor? Is this an i815E board or a Via board?

I thought it had itnernal diode support.... but if it has a socket-thermistor, it probably can't read the internal diode at all then



Mike

 

[Duallys]

Yep ... the GA-6OXET is an i815EP based board with a socket thermistor. So, no joy eh?

 

[Mikewarrior2]

its not often that you see an intel chipset MB with a sokcet-thermistor, but because it is there, you're likely not going to able to find a way to read teh internal diode



Mike

 

[Rack]

I grew impatient with the lack of AMD XP diode support, so I just rolled my own:

one of my posts on my adventures at Ars

Now if I just had an XP chip to attach it to... (it's in the mail).

I'll start testing heatsinks myself and then I can laugh at results like the ones that HardOCP puts out.

 

[Emo]

It looks to me that the question of socket thermistor raised by the AS person would become moot if one would test different thermal compounds on a P4, right? That would eliminate the external sensor inaccuracy. I can't believe that no hardware website has done it! IMHO, while the socket thermistors may not be very accurate, given the same cpu/mobo/heatsink setup and changing only one factor (thermal compound) should give us at least a rough idea of the AS performance. It would be really interesting to test the thermal compounds on a board with both a socket thermistor and the P4 internal diode and compare the readings. I imagine that would put a period somewhere (be it AS or the nay-sayers).

 

[Mikewarrior2]

Rack,

Eagerly looking forward to your results.... hope they work well.


Emo,

Even if you can balance out an exact system with a socket-thermistor setup to test AS and other greases, you get these issues: Compression, to the order of at least 1/2, but more like 1/3-1/4. That means, for ever 1C the core temp changes, the socket-thermistor will register from .5-.25C. OR even less. And since MBM monitors measure changes in 1C increments, you're looking at most changes not showing up at all. You also have the issue of thermal grease contamination. I've seen plenty of grease reviews where they barely take the time to clean the top of the CPU to try to remove as much of the old grease as possible.

The result is this: You will get a temp drop with AS, but it will not be shown as the full temp drop. I believe that was the main issue with Nevin's interview, since you've got a lot of websites pranting and ranting about how AS doesn't really drop temps over normal greases.

A cool, new example is the Cluboverclocker test between Arctic Alumina and Arctic Silver 2. A Celeron with Internal Diode showed a 4F drop. A Socket-thermistor T-bird showed a 3F drop. Since the T-bird was more than double the wattage of the Celeron, one would expect an over double drop with AS versus AA. So in there review, its clear that the Socket-thermistor is compressing the results by at least a factor of 3.

The problem with using a p4 is this: The core is massive compared to a t-bird/p3. At least hte p3/t-bird/athlon XP were simliar in size. Even if you take off the IHS, the P4 is some ~220 sq mm and the XP is ~128 sq mm. The comparison would work, but the results wouldn't quite be the same, since the smaller contact surface and higher thermal density of the XP making cooling harder.



Mike

 

[Nevin]

Emo,

While there have not been any thermal compound tests published that used the P4 internal diode, there have been quite a few using the P3 or Celeron II internal diode. As Mikewarrior said, these chips with their smaller contact areas (Similar to AMD chips) are better interface material test platforms than the P4. If a PIII and a P4 were dissipating the exact same power, the PIII would show almost 6X the difference between any two thermal compounds than the P4 would show as the PIII's contact area is only 1/6th that of the P4. Since the P4s are actually dissipating 2 to 3 time the power of the PIIIs however, the difference in a real test would be 1/2 to 1/3 the difference shown on a PIII.

This is a good time to consider that the difference between 2 thermal interfaces on a high-performance AMD CPU will be at least double the difference on a PIII since the contact area is similar and the AMD chip dissipates at least double the power.

To put all of this in perspective, let's say that the...

Internal diode measured difference on a PIII between 2 TIMs = 4C

Then...
We know that the internal diode measured difference between the same two TIMs on a P4 dissipating 3 times as much power will = 2C. (4 / 6) X 3.

And...
The ACTUAL internal core temperature difference on an AMD CPU dissipating 2 times the power will = 8C. (4 x 2.) (But, we also know that there is no way to measure it.)

Here are six published tests using the internal diode on an Intel CPU to evaluate thermal interface material performance:

The Tech Zone Thermal Compound Shootout

Target PC

Club Overclocker

Maximum 3D

The Hardware Pub (This site was down at the time I posted this, but hopefully will be back up soon.)

Tweakers AS2 Test
Tweakers Thermal Compound Shootout
Note: You need to look at both Tweakers tests if you want a complete comparison. (They are in German, but the graphs are easy to understand.)


And here is an arcticle at Overclockers.com comparing internal diode measurements to in-socket thermistor measurements using an Intel CPU on a modified slotket.

Intel Internal Diode vs. In-Socket CPU Temp Comparison

Nevin House
Arctic Silver, Inc.

 

[Rack]

Looks like I'll have to do a thermal compound test while I'm at it.
But I still haven't come up with a good testing protocol that wouldn't be biased by the pastes tested before and still be have a large drop across the paste. Lapping would remove the bias, but it would close the gaps, making the resistance drop less. I'll just have to think about it more. So far, I just have "use more voltage and MHz" to get a bigger spread. Maybe "use mcx462 with 80cfm/80mm delta" to minimize the sink resistance, or maybe just a retail sink, I haven't thought it out yet.
Maybe a thermocouple in whatever heatsink would be in order, so I could have data points above and below the paste on the thermal gradient.

 

[Atif]

In case you're interested, VN Roundup is currently giving away three tubes of the as-of-today unreleased Arctic Silver III Thermal compound thanks in 100% to Arctic Silver having provided it to us. If you guys are interested, just drop by and enter to win one.

Peace