NVBleed: Covert and Side-Channel Attacks on NVIDIA Multi-GPU Interconnect
Yicheng Zhang, Ravan Nazaraliyev, Sankha Baran Dutta, Andres Marquez, Kevin Barker, Nael Abu-Ghazaleh · 2025
Read NVBleed: Covert and Side-Channel Attacks on NVIDIA Multi-GPU Interconnect by Yicheng Zhang, Ravan Nazaraliyev, Sankha Baran Dutta, Andres Marquez, Kevin Barker, Nael Abu-Ghazaleh (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-17847.txt.
Description
Yicheng Zhang and co-authors show that the fast links between graphics processors in big computing clusters can leak secrets. Multi-GPU machines use NVIDIA's NVLink to move data quickly, and the team reverse engineered how that link behaves. They found two leak sources: timing shifts when traffic collides, and built-in counters that reveal communication patterns.
From those leaks they built hidden channels between two processors carrying over 70 kilobits per second with under five percent errors. They also ran spying attacks that identified which program was running, across eighteen test applications, and read characters from a 3D rendering program. The leaks even crossed between separate cloud accounts on Google's platform. The paper closes by weighing defenses.
Catalog record
| CREATOR | Yicheng Zhang, Ravan Nazaraliyev, Sankha Baran Dutta, Andres Marquez, Kevin Barker, Nael Abu-Ghazaleh |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-17847