Sustainable High-Performance Computing: Assessing Environmental Impacts and Future Strategies

In this study, we developed a multi-layered model analyzing HPC energy use, carbon footprint, and broader environmental consequences. We built a hierarchical model that categorizes energy consumption based on system scale. Our calculations indicated that global HPC energy consumption in 2023 was 82.09 TWh. We used a weighted carbon intensity formula to estimate carbon emissions using forecasted energy consumption and national energy mixes. The model considered both the manufacturing and operational phases of HPC systems. In 2023, the carbon emissions from HPC systems were approximately 35 million tons of CO₂, primarily due to fossil fuel energy consumption. We used feature engineering techniques to transform the time series forecasts of HPC energy consumption and energy mixes into regression problems. We applied Gradient Boosting (CatBoost) for iterative forecasting. As renewable energy sources become a more significant part of the global energy mix, the renewable share in HPC energy use is expected to grow from 30% in 2023 to 46.9% by 2030. This shift will help reduce carbon emissions, although HPC energy consumption is projected to increase to 86.56 TWh by 2030, showing a consistent linear growth trend. We also developed a water consumption model for HPC cooling systems. It underscores the significant impact of energy structure optimization on emission reductions.