Reproduce “Bit2Watt: A Cyber-Physical Vulnerability Exploiting GPU Workloads Across Power and Computing Infrastructures” 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
Zhouhao Ji, Kaikai Pan, Wenyuan Xu
Researchers from Zhejiang University identify and analyze "Bit2Watt," a cyberphysical vulnerability where legitimate GPU workloads can induce high-frequency power modulations. These modulations degrade local power quality, destabilize inverter-dominated grids, and can create a "Watt2Bit" feedback loop that disrupts computing services and enables covert data exfiltration, demonstrating power quality degradation (e.g., THD up to 46.8%) and system instability (e.g., negative damping ratio of -0.27).
Key insights
- GPU power draw can be precisely and rapidly modulated at high frequencies (up to 6 kHz) through legitimate software workloads, acting as a controllable excitation source for the power grid.
- The constant power load (CPL) characteristics and negative incremental resistance (NIR) of GPU clusters, combined with the resonant frequency responses of inverter-based DERs, create a positive feedback loop that can lead to significant power quality degradation and system instability.
- Power disturbances induced by GPUs can propagate back to disrupt computing services (Watt2Bit), causing denial-of-service via thermal stress and protection trips, and enabling covert information exfiltration through electromagnetic interference (EMI) side channels.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.