A simulated die has the advantage of having no secondary cooling paths (if properly set up) and having a way to control and measure pretty exactly the heat load produced. As you move towards studying 0.1C or 0.01C differences then knowing the heat load precisely becomes critical (and impossible on actual systems). That was why I said that CPUs could never be used for high end testing like BillA and Joe are talking about.
Another good point though: secondary cooling paths can be substantial. I talked with Andrew at Millenium Thermal about this (he worked on the Agilent Articooler also) because I was getting far better temps on Dynatron heatsinks with 44CFM Delta fans than the Overclockers die would produce. Other heatsinks seemed to fall pretty much in line with Joe C's testing. In the end, I realized that with really high CFM, high pressure fans that secondary cooling might be important. This would be kinda interesting to study. You might want to come up with an adjustable benchtop PSU for the fans though; power supplies on computers will have variable voltage.
When all is said and done, the parameter that frightens me is paste application. Ingenuity can deal with changes in pressure, suitcases of money would solve the resolution/accuracy issues (and purchase a very nice simulation setup), but how to deal with paste application? What about differences in consistency as the tube of paste ages? I have an old Arctic Alumina tube that seems to have gone bad. What about changes in formulation from batch to batch of paste? How many replicates do I need to do to capture this variability: 5, 10, 100?
|