Hi NVIDIA Developers,
I’m posting on behalf of Confident Security, a private AI inference company. We’re open-sourcing go-nvtrust, inspired by NVIDIA’s nvtrust, for NVIDIA GPU and NVSwitch confidential computing attestation. This library provides a Go implementation for collecting attestation evidence from Hopper (H100) and Blackwell GPUs, as well as LS10 NVSwitches, with verification through the NVIDIA Remote Attestation Service (NRAS).
Key features include:
• Go bindings for libnvidia-nscq (NVSwitch attestation library)
• Integration with NVML via go-nvml for GPU attestation
• Simple API for evidence collection, remote verification, and confidential computing workflows
It’s designed to help developers build secure AI inference systems by ensuring hardware trust against threats like insiders or nation-states.
Repository: GitHub - confidentsecurity/go-nvtrust: go version of https://github.com/nvidia/nvtrust
License: Apache-2.0
Quick start and API reference in the README.
We’d appreciate any feedback, contributions, or discussions on how this fits into your confidential computing setups. Has anyone here implemented GPU/NVSwitch attestation in Go, or have suggestions for improvements?
Cheers,
Sam