PowerDL: A unified in-memory GPU energy profiling and visualization toolkit for deep learning

Sancar Y.

SOFTWAREX, cilt.34, 2026 (SCI-Expanded, Scopus) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 34
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1016/j.softx.2026.102627
  • Dergi Adı: SOFTWAREX
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Directory of Open Access Journals
  • Atatürk Üniversitesi Adresli: Evet

Özet

This study presents PowerDL, an in-memory software tool designed for deep learning workloads that profiles GPU energy consumption, power usage, and hardware utilization rates. PowerDL provides a unified interface for PyTorch and TensorFlow frameworks, enabling high-time-resolution analysis of GPU behavior during training and inference phases. The software supports interactive exploration of energy and performance metrics without requiring mandatory file writing. Time-series-based power traces, cumulative energy consumption, and phaseaware visualizations enable users to conduct energy-aware deep learning experiments. Additionally, optional output export mechanisms are provided to support reproducible research.