GPUHammer: Rowhammer Attacks on GPU Memories are Practical
Chris S. Lin
34th USENIX Security Symposium (USENIX Security '25) · Day 3 · Hardware Security 3: Side-Channel and Fault Injection Attacks
The talk "GPUHammer: Rowhammer Attacks on GPU Memories are Practical" presents a groundbreaking study demonstrating the first practical **Rowhammer** attacks specifically targeting Graphics Processing Unit (GPU) memories. Presented by Chris Joyce, a researcher from the University of Toronto, this work highlights a critical security vulnerability in a hardware component increasingly central to modern computing, particularly in machine learning (ML) and high-performance computing. The research underscores that GPUs, much like CPUs, are susceptible to Rowhammer-induced bit flips, posing a significant threat to data integrity and the security of sensitive applications.
AI review
Solid, well-scoped hardware security research that extends Rowhammer — a mature but still-evolving attack class — into genuinely underexplored territory: GPU DRAM. The three-challenge framework (address mapping, hammering intensity, TRR bypass) is methodically solved with novel adaptations, and the ML accuracy-degradation demo gives the work real-world bite beyond a pure academic exercise.