Answers are grounded in this paper and cite their sources.
Systematic mapping study identifies key environmental impacts of high-performance computing, suggesting major gaps.
High-performance computing underpins climate modeling, drug discovery, and large-scale artificial intelligence, yet its environmental footprint reaches beyond operational energy to embodied carbon, water consumption, and electronic waste. Knowledge of these impacts remains scattered across fragmented studies, making it hard to see what is well understood and what stays overlooked. This systematic mapping study charted the research landscape on the environmental impacts of high-performance computing to identify dominant themes, recurring methods, and structural gaps. We searched IEEE Xplore, ACM Digital Library, and Scopus through April 2026, retrieving 4812 records that reduced to 3788 after deduplication. Title and abstract screening combined manual review with three large language model assessors and backward and forward snowballing, yielding a final corpus of 62 reports. A keywording process produced a four-facet classification covering environmental impacts, lifecycle stages, system locus, and management and intervention levers. Carbon emissions appeared in 54 of the 62 reports, while materials and waste, water use, and biodiversity received much sparser coverage, and most studies addressed a single impact in isolation. Activity concentrated on the operational stage and on facility and energy infrastructure, leaving water beyond cooling, biodiversity, and end of life processes largely unmeasured. Reported efficiency improvements were repeatedly absorbed by growing demand rather than translating into absolute reductions. The map suggests that efficiency-focused research alone cannot deliver environmental sustainability for high-performance computing. Progress may depend on building instrumentation for under-measured impacts, treating lifecycle cost as a procurement criterion, and adopting sufficiency-aware accounting that links environmental cost to delivered scientific value.
No takes yet. Share an insight, caveat, or question.
Rosa et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: