Abstract: To take effective management decisions, both power and thermal management of multicores depend on accurate run-time dynamic power consumption information at core-level. Due to the cost-prohibitive nature of actually measuring core power, such run-time power information is usually derived from predetermined power models which use observable performance counters, operating frequency and voltage as inputs. The performance counters are necessary to model the activity and thus indirectly the power consumption of each core. Deriving such power information is mainly based on predetermined power models which use linear modeling techniques to determine the core-performance/core-power relationship. However, with multicore processors becoming ever more complex, linear modeling techniques cannot capture all possible core-performance related power states anymore. In this talk, we present different statistical and stochastic (including neural networks) methodologies that can be used to better model power consumption at core-level, hence allowing to improve power and thermal management for more sustainable embedded systems.