E11 - GPUHammer Exposé: Alex Breaks Down the First RowHammer GPU Attack

Join Alex on this episode of IT Leadership for Small Business as he dives into the groundbreaking GPUHammer research—the first-ever RowHammer-style exploit targeting NVIDIA GPUs. Alex unpacks:

• How GPUHammer manipulates GDDR6 memory (e.g., on NVIDIA A6000) to flip bits and derail AI models, slashing accuracy from 80% to near-zero 

• The fundamental mechanics of RowHammer and why it now impacts GPUs as well as CPUs

• The significance for AI-driven apps in shared GPU environments, and implications for cloud platforms 

• Recommended mitigations: enabling system-level ECC via nvidia‑smi ‑e 1, at the cost of ~10% inference slowdown—and why newer NVIDIA chips like H100/RTX 5090 are immune due to on‑die ECC

Tune in for a clear, jargon-free breakdown of this hardware-level threat—and what steps IT leaders should take now to safeguard GPU-accelerated workflows.