A Comparative Study of Model Training Time and Cost on AWS, Azure, and GCP using DevOps Automation

2025/06/05 by Okorie, Joshua
#Computational Engineering #Engineering #cloud computing #infrastructure as code #machine learning #model training #performance benchmarking

paper · doi:10.17605/osf.io/48jzd

Abstract

Selecting the best cloud platform for model training, balancing cost, scalability, and performance, becomes essential as businesses use machine learning (ML) solutions more and more. This study compares the costs and duration of model training for three of the top cloud service providers: Google Cloud Platform (GCP), Microsoft Azure, and Amazon Web Services (AWS). I deployed and trained identical ML models on each platform using a consistent DevOps pipeline driven by Terraform, Docker, and GitLab CI/CD. For CPU and GPU-based environments, the study assessed compute performance, training time, resource provisioning time, and billing efficiency. Additionally, I examined hidden expenses related to data transfer, storage, and idle computation. Data scientists, DevOps and MLOps engineers, and technical decision-makers can use this reproducible benchmark to help them choose the most economical and performant cloud infrastructure for ML workloads. The code and infrastructure scripts for this research are publicly available at: https://github.com/stephenjoshua18-web/Comparative_Study_DevOps_Automation

Related

• A Comparative Study of Model Training Time and Cost on AWS, Azure, and GCP using DevOps Automation
• Deep Learning Model Deployment in Multiple Cloud Providers: an Exploratory Study Using Low Computing Power Environments
• ContainerStress: Autonomous Cloud-Node Scoping Framework for Big-Data ML Use Cases
• How Can We Train Deep Learning Models Across Clouds and Continents? An Experimental Study
• Machine Learning and Cloud Computing: Survey of Distributed and SaaS Solutions