EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Distributed Training Strategies for Reducing Carbon Footprint in Large Scale Model Development

Yuxuan Cheng, Zekai Sun, Michael Turner

Computer Life · 18.05.2026 · 14 (2) · 8-15

DOI: 10.54097/yhppk428 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer Life, online erschienen am 18.05.2026. Verfasst von Yuxuan Cheng und 2 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid expansion of deep learning has led to increasingly large-scale neural networks whose training demands massive computational resources, resulting in substantial carbon dioxide (CO2) emissions. This paper investigates distributed training strategies specifically designed to reduce the carbon footprint associated with large-scale model development. We propose an Energy-Aware Distributed Training (EADT) framework that integrates gradient compression, mixed-precision arithmetic, and carbon-conscious workload scheduling across heterogeneous graphics processing unit (GPU) clusters. By combining ring-All Reduce communication protocols with adaptive sparsification and low-precision quantization, the proposed framework reduces both inter-node communication overhead and total floating-point operations, thereby lowering energy consumption. Experimental results demonstrate that the EADT framework achieves up to a 40.0% reduction in estimated CO2 emissions compared to baseline full-precision data-parallel training, while incurring only marginal losses in convergence quality. These findings highlight the potential of communication-efficient and computationally frugal distributed training paradigms as practical tools for greener artificial intelligence (AI) development.

Zitieren

APA

Cheng, Y., Sun, Z., & Turner, M. (2026). Distributed Training Strategies for Reducing Carbon Footprint in Large Scale Model Development. Computer Life, 14(2), 8-15. https://doi.org/10.54097/yhppk428

Vancouver

Cheng Y, Sun Z, Turner M. Distributed Training Strategies for Reducing Carbon Footprint in Large Scale Model Development. Computer Life. 2026;14(2):8-15. doi:10.54097/yhppk428

BibTeX
@article{Cheng2026,
  author = {Cheng, Yuxuan and Sun, Zekai and Turner, Michael},
  title = {Distributed Training Strategies for Reducing Carbon Footprint in Large Scale Model Development},
  journal = {Computer Life},
  year = {2026},
  volume = {14},
  number = {2},
  pages = {8--15},
  doi = {10.54097/yhppk428}
}
Kostenlos starten →