divagr18 / gpu-instance-fingerprint
git clone https://github.com/divagr18/gpu-instance-fingerprint.git
Fingerprint the exact GPU you're running on, even among identical GPUs, from the order its atomic adds land in. 99.1% accuracy, survives reboots, can't be faked, and a default torch.index_add_ call leaks it.
+2
stars, 24h · #15,966
+29
stars, 7 days · #9,015
+29
stars, 30 days · #23,934
29
★ total · +29% this week
5
forks · 0 open issues
Star history (as recorded here)
LanguagePython
LicenseMIT
Created2026-10-30 (4d ago)
Last push4d ago
OwnerUser
GitHub topics
cudacupydevice-fingerprintingfingerprintingfloating-pointgpuhardware-securitynondeterminismnvidiaprivacypytorchside-channel
Links
- GitHub: github.com/divagr18/gpu-instance-fingerprint
- Homepage: https://divagr.com/blog/telling-identical-gpus-apart
- Releases · Issues · Hacker News mentions
README
Loading README…