--- format: "aidr-story-markdown/v1" id: "caf8646d63fa0c34d3ba42093149e9ded609e4635116b4aa05c450fd004e2462" canonical_url: "https://aidr.today/caf8646d?lang=en" title: "Standard PyTorch Call Enables Unique Fingerprinting of Identical GPU Hardware" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-10-31T19:08:00.000Z" category: "Frameworks" topics: ["coding","framework"] source_urls: ["https://huggingnews.com/tech/standard-pytorch-call-enables-unique-fingerprinting-of-identical-gpu-har-7609fe22","https://x.com/divagr1925/status/2116241086050115692","https://x.com/giffmana/status/2116599971176935519","https://marketbrief.now/tech/standard-pytorch-call-enables-unique-fingerprinting-of-identical-gpu-har-7609fe22"] summary: "A PyTorch operation called torch.index_add_ can extract a unique, persistent hardware identifier from individual GPUs, even when they are the same model. The method relies on variations in floating-point math results caused by differences in physical GPU hardware, and the identifier reportedly survives system reboots and cannot be spoofed." --- # Standard PyTorch Call Enables Unique Fingerprinting of Identical GPU Hardware > [Open the canonical story]() **Published:** 2026-10-31T19:08:00.000Z **Category:** Frameworks **Topics:** coding, framework ## Summary A PyTorch operation called torch\.index\_add\_ can extract a unique, persistent hardware identifier from individual GPUs, even when they are the same model\. The method relies on variations in floating\-point math results caused by differences in physical GPU hardware, and the identifier reportedly survives system reboots and cannot be spoofed\. ## Sources - [Story source]() - [Story source]() - [Supporting source]() - [Story source]()