DVS® technology allows the rack node energy estimation. The estimates are used by an optimization routine that defines step by step the best actions to perform (node switch off policy, thermal aware job scheduling policy, …) to find the best trade-off between energy consumption and the quickest jobs execution. In parallel the HPC node thermal evolution is predicted by a second DVS®, and a trade-off between HPC cooling and energy saving policy is guaranteed by the implementation of an STC® control.