Another thought. I'm beginning to wonder how linear the CPU temp diodes, and/or Coretemp is performing. I will attempt to better control my ambient variation when I take test points.
Great job you put a lot of effort into this test much appreciated.
The film impressions are great and making me think. The initial print had the loads moat wise center then edge. The second print has the loads redistributed inboard from the edge but less pressure on center so may have a kind of averaging effect.
Sooooo......... If you have good contact at the start on center maybe on this moat kind of a profile you just take a flat bastard file to to the edges to increase pressure in the center region only and leave the center region untouched. Maybe relieving as much to leave the only contact area over the die as the only possible contact area with the sink to even further concentrate pressure on center
Another option is you may be at be at an optimum contact with the die from the start so any changes could be would be merely minimal.
In my experience when I make large changes as you did with the lapping and see no or minimal changes there is usually a third dominant factor somewhere that is masking result.
It's good you ran the test with ICD7 before and after. The Idle vs load readings in coretemp are totally at odds with with physics.
ICD7 contains 92% diamond some carbon black and a couple % synthetic oil. All the materials are stable through the range of heat you guys use with a quadruple margin. It will not change in thermal resistance across the range of temps. The same holds true for any compound I ever tested, C/W is C/W it does not change it is a constant we can now use
So you have a 2C margin improvement@ idle, say your are running 50W in idle (I do not know your idle Watts were just approximating) with a TIM resistance on the Die of .01-.02 C/W At 100W the margin can only increase (C/W is C/W) the spread on temps would now be in the range of 2-4C @ 150W 3-6C @ 200W 4-8C.
Resolution on your test gets better the higher you go. Your original interface resistance before lapping it dropped 1-2C and was probably in the 02-.04 C/W range so on idle you have approx. a 50% improvement on the thermal interface which would hold through the range
Your numbers are actually going in the opposite direction which is contrary to anything in my experience as how a thermal grease performs. I've seen this on a number of those in the marginal improvement category.
I would suspect that it would be more likely subtle variations in the diodes as I have both great and marginal results on all software programs which is a great call on your part.
Overall my bet would be that if you had a thermocouple on the set up you would have seen a 4-5C drop overall. Testing is hard work, thanks
Edit: Compare load and Idle tests on this test before and after. Idle dropped by a significant amount and load remains unchanged The delta temp can only go up with increased watts
http://img528.imageshack.us/img528/9...ltsimagwv4.png
Reply With Quote
Bookmarks