Single-Chip Cloud Computer thermal model

Spatial and temporal non-uniformities of workload and power consumption advanced Systems-on-Chip (SoC) platforms result in localized high power densities, which lead to temperature hot-spots, gradients and thermal cycles that may cause non-uniform ageing and accelerated chip failure. The Single-Chip Cloud Computer (SCC) is an experimental many-core processor created by Intel Labs and it integrates thermal sensors to track the chip thermal behavior. Unfortunately these sensors provide a limited introspection on the full-chip thermal map, as they monitor temperature at a coarse granularity. In this paper we build a fine-grained thermal and power model of SCC using a state-of-the-art thermal modeling tool (Hotspot). We calibrate the model with measured data from chip sensors. We assess the predictive power of the thermal model and its main sources of error.