GAIA: GPU Power Attacks for AI Data Centers
GAIA: GPU Microarchitecture-Based Power Attacks and Countermeasures for AI Data Centers
Project Summary
AI data centers rely heavily on GPU clusters for training and inference of large models. This project investigates microarchitecture-level power side-channel attacks targeting GPU hardware in AI data center environments, and develops corresponding countermeasures.
The GAIA project aims to:
- Characterize power consumption patterns exposed by GPU microarchitectural features during ML workloads
- Develop practical side-channel attack methodologies that can extract sensitive model weights or intermediate data
- Design and evaluate hardware and software countermeasures to mitigate these vulnerabilities
Research Team
- PI: Dr. Arafin, George Mason University
- Co-PI: Khaled Khasawneh, George Mason University
Funding
This project is supported by the Virginia Innovation Partnership Authority (Award #N-1Q27-004).