The Importance of Heat-Sink Modeling for DTM and a Correction to “Predictive DTM for Multimedia Applications”

It is anticipated that in the future, increasing power densities and resultant temperatures will often be the dominant limit to processor performance and a significant component of cost. Dynamic thermal management (DTM) techniques address this issue by allowing the thermal solution to be designed for a temperature less than the peak. Evaluating DTM and other processor thermal properties requires reliable tools for processor thermal modeling. In particular, the processor heat-sink model plays a crucial role, but much prior work in DTM, including our own, is based on simplistic assumptions of heat-sink behavior. Using more realistic heat sink simulation, this paper corrects our previous work and underscores the need for more accurate heat sink simulation for future thermal work. Specifically, we re-examine the DTM results in our paper, ”Predictive Dynamic Thermal Management for Multimedia Applications” [12]. We show that although the predictive DTM algorithms discussed in that paper are still effective, incorrect heat sink simulation resulted in erroneous conclusions on the effectiveness of global vs. local response mechanisms. We also describe how for reactive DTM algorithms (which constitute all of the prior DTM work), heat sink simulation is more complex than previously discussed.