Modeling of energy consumption and associated carbon emissions for HPC systems

As the demand for AI innovations and large-scale data processing increases, so do the energy consumption and carbon emissions of HPC systems. These systems require significant energy for computation and cooling, making them major contributors to global energy use and carbon emissions. Over time, we introduced a growth model for HPC systems. By 2024, with average, complete, and peak utilization, energy consumption was estimated at approximately 13,202.4 GWh, 16,299.3 GWh, and 25,012.5 GWh, respectively. We gathered data on the energy mix from 1985 to 2023 and established a carbon emissions model to analyze the results. In 2024, considering average, complete, and peak utilization, the carbon emissions were estimated to be approximately 6.7202 TgCO2eq, 8.2965 TgCO2eq, and 12.7317 TgCO2eq, respectively. We established a set of models to simulate the growth of HPC and the potential impacts of various energy sources and mixes. We employed curve fitting for the eight energy sources by selecting key years based on their trend shapes, achieving an average R2 of 0.9426, indicating a strong model fit. We also presented a model to simulate water consumption. By 2030, considering average, complete, and peak utilization of the HPC systems, water consumption will be approximately 80.1265 GL, 95.707 GL, and 153.314 GL, respectively.