University of Toronto researchers introduced GPUThor, a Rowhammer attack using non-uniform hammering against error-correcting-code-protected NVIDIA GPUs.

University of Toronto researchers published “GPUThor: Amplifying Rowhammer Attacks via Non-Uniform Patterns to Exploit ECC-Protected GPUs.” The abstract describes GPUThor as a high-intensity Rowhammer attack on NVIDIA graphics processing units (GPUs), including devices protected by error-correcting code (ECC).

The stated technical approach reverse engineers GPU memory-access coalescing behavior to enable non-uniform hammering patterns. The excerpt identifies activation of aggressor rows as part of the mechanism, but stops before providing the complete experimental description.

The research concerns GPU memory security, not a newly reported compromise of a named cloud provider or production cluster. The supplied material does not identify affected NVIDIA models, attack success rates, required access privileges, or available mitigations, and therefore does not establish that every ECC-protected NVIDIA GPU is vulnerable.