reports on a novel side-channel attack
targeted at multiple graphic processing unit-dependent architectures that was developed by researchers from the University of California, Pacific Northwest National Laboratory, and Binghamton University, as well as an independent contributor, by reverse-engineering the cache hierarchy of NVIDIA's Pascal-based DGX-1 system.
Cache hierarchy reverse-engineering has enabled attackers on one GPU to trigger cache contention on the other GPU. Researchers also discovered that reverse-engineering cache sharing has allowed one GPU to remotely access others' caches.
Moreover, proof-of-concept side-channel attacks developed by the researchers allowed the recovery of remote victims' access memorygram, which was then leveraged for application fingerprinting on the victim GPU.
"This attack can be used to identify and reverse engineer the scheduling of applications on a multi-GPU system (simply by spying on all other GPUs in a GPU-box), and identify a target GPUs that are running a specific victim application, and even identify the kernels running on each GPU," said researchers.