Related Experiment Videos
Measuring the impact of virtualization and containerization on the environment when using GPUs for processing the AI
Safaa Hriez1, Mohammad Haikal2
1Cyber Security Department, Al Hussein Technical University, Amman, Jordan.
Frontiers in Big Data
|July 1, 2026
Summary
Containerization is more energy-efficient than virtualization for AI workloads. This approach reduces energy consumption and carbon dioxide emissions by approximately 21.6% compared to virtual machines.
Area of Science:
- Computer Science
- Artificial Intelligence
- Sustainable Computing
Background:
- Artificial intelligence (AI) growth drives high computing demand.
- Virtualization and containerization optimize resources but their energy efficiency for AI workloads is unclear.
Purpose of the Study:
- To compare the energy efficiency and environmental impact of containerization versus virtualization for GPU-accelerated AI workloads.
- To quantify differences in power consumption and resource utilization.
Main Methods:
- Trained a DenseNet-121 model on MNIST using VirtualBox and Docker.
- Measured performance, GPU utilization, and power consumption for both environments.
Main Results:
- Containerization (Docker) reduced energy consumption by ~21.6% compared to virtualization (VirtualBox).
- Containerization showed lower average and peak GPU utilization and power draw.
- Significant reductions in carbon dioxide emissions were observed with containerization.
Conclusions:
- Containerization is a more energy-efficient and sustainable choice for GPU-enabled AI workloads.
- Results suggest potential for reduced environmental impact without performance loss.
- Further research is needed for diverse AI models and hardware.
