In today's fast-paced electronic product iterations, many powerful chips are abandoned because they are "not top-tier," leading to significant resource waste. DGrid proposes a highly socially responsible solution: through algorithm optimization, these "sub-top" hardware are regrouped. In DGrid's distributed architecture, the collaborative effect of hundreds of mid-range GPUs can easily handle inference tasks that originally required top-tier accelerators.