How Carbon Metrics Impact Device Selection

As computing systems increasingly contribute to carbon emissions, understanding how comprehensive carbon metrics influence device selection is crucial for sustainable computing. We investigate how considering embodied carbon alongside operational carbon affects optimal device choice for AI inference workloads. Our results show that including embodied carbon changes the optimal device choice for up to 58% of workloads, with the impact being more pronounced in low carbon intensity regions. This demonstrates that operational carbon alone is insufficient for sustainable device selection, highlighting the need for comprehensive carbon-aware metrics.