#memoryattack

[cyber-sec]

New Attack Wrecks ML Models on NVIDIA GPUs

Hackers can now flip bits in NVIDIA A6000 GPU memory using a Rowhammer-style attack called GPUHammer, sabotaging deep learning models with a single hit. The University of Toronto's research shows how even advanced GPU memory is vulnerable.

Source: GPUHammer

Read more: CyberSecBrief