The problem
Can a scheduler choose the lowest-energy placement if the model is incomplete or the hardware is changing underneath it?
What current research shows
Linux EAS predicts the energy impact of CPU placement using capacity and energy models. The kernel documentation states that current EAS is SMT-unaware, which limits its ability to reason about multithreaded hardware. The model is a decision aid, not a direct measurement of package energy.
Where the evidence stops
A model can be directionally useful while still missing firmware behavior, shared resources, thermal limits, device work, and memory energy.
What Valen Systems is testing
Pair every energy-aware decision with measured work, package energy, temperature, and tail latency on the disclosed hardware.