Our Investigation is the initial systematic analyze of on-system LLM execution, quantifying efficiency, Electrical power efficiency and accuracy throughout numerous condition-of-the-artwork designs and showcases the condition of on-device intelligence while in the period of hyperscale versions. Effects highlight the functionality heterogeneity thro