Reproduce “GPU Fingerprinting for Location Verification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wayne Tee, Jonathan Happel
The authors propose GPU hardware fingerprints to identify a responding chip in a location-verification scheme that separately uses response latency to constrain distance. Their function uses server-selected seeds, and fingerprints are matched by L1 distance. Tests on 24 Nvidia H200 GPUs achieved 98.8% accuracy with one verification run and 100.0% with two, demonstrating small-scale identification, not end-to-end location verification.
Key insights
- Microscopic manufacturing variations in individual GPU chips create unique, uncloneable hardware characteristics that can serve as reliable fingerprints.
- Race conditions within atomic operations on GPUs can be systematically exploited and measured to generate distinct, verifiable hardware fingerprints.
- Parameterizing the fingerprinting function with random seeds is crucial for preventing replay attacks and ensuring that each verification challenge produces a unique, unpredictable response from the physical GPU.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.