Confidential Computing within an AI Accelerator
summary
Built an AI accelerator extension called IPU Trusted Extensions (ITX) that adds an on‑chip hardware root-of-trust and PCIe Gen4‑rate cryptographic engines, plus compiler/runtime changes to support multi‑party training without a CPU TEE. Implemented on Graphcore GC200 taped out at TSMC 7nm and tested on a dev board, it keeps models and data encrypted on the device, decrypts at chip boundary, and reports <5% overhead in typical DNN training and up to 17x speedups versus CPU SEV‑SNP solutions.
AI-written summary · about
chapters (8)
- 0:00Confidential computing in AI
- 1:07Why confidential computing for AI cloud
- 3:15Low-TCB AI accelerator design
- 8:49Trust extensions for IPU
- 13:44Remote attestation and key provisioning
- 16:35Replay and reorder attack protection
- 17:19Performance overheads and benchmarks
- 19:09Future of confidential AI
related talks
42mSponsored Session: Confidential Computing - New Capabilities for New Workloads
42mSponsored Session: Confidential Computing - New Capabilities for New Workloads
37mConfidential AI: Enabling Trustworthy Ecosystems in the Age of Agentic Intelligence - Session
20mTrusted Hardware for AI Workloads: Extending Confidential...
44mGPU-less, Trust-less, Limit-less: Reimagining the Confidential AI Cloud
36mSecurity for AI and Multi-Party Collaboration with Confidential Computing (Oh, and...
20mBuilding Trusted AI Factories with NVIDIA Accelerated Confidential Computing
20mKeynote: Future-Proofing Confidential AI
