Characterizing the Efficiency of Distributed Training: A Power, Performance, and Thermal Perspective

Go, Seokjin; Park, Joongun; More, Spandan; Wu, Hanjiang; Wang, Irene; Jezghani, Aaron; Krishna, Tushar; Mahajan, Divya

doi:10.5281/zenodo.16734738

Published October 18, 2025 | Version v0.1

Software Open

Characterizing the Efficiency of Distributed Training: A Power, Performance, and Thermal Perspective

1. Georgia Institute of Technology

This artifact includes the source code for the paper "Characterizing the Efficiency of Distributed Training: A Power, Performance, and Thermal Perspective" published in the MICRO 2025 conference. It contains the evaluated frameworks and a modified Zeus implementation for hardware telemetry collection. It also provides SLURM and bash scripts to prepare and launch experiments, along with Python scripts to reproduce and visualize key results from the paper.

Files

Files (68.6 MB)

Name	Size	Download all
aiinfra-artifact.tar.gz md5:d2f8e4b0e0ec3ade88fc52c85dcbf586	68.6 MB	Download

Views

Downloads

Show more details

	All versions	This version
Views	82	82
Downloads	16	16
Data volume	1.2 GB	1.2 GB

More info on how stats are collected....

DOI

Resource type

Software

Publisher

Zenodo

License: MIT License

A short and simple permissive license with conditions only requiring preservation of copyright and license notices. Licensed works, modifications, and larger works may be distributed under different terms and without source code. Read more

Technical metadata

Created: August 4, 2025
Modified: August 4, 2025

Characterizing the Efficiency of Distributed Training: A Power, Performance, and Thermal Perspective

Creators

Description

Files

Files (68.6 MB)