Show HN: Go-nvtrust – Go library for Nvidia GPU attestation (github.com)

go-nvtrust is a new Go library that implements NVIDIA GPU and NVSwitch confidential-computing attestation (inspired by nvidia/nvtrust). It provides Go bindings to NVML (for Hopper H100 and Blackwell GPUs) and libnvidia-nscq (for NVSwitch LS10), collects attestation evidence, sends it to NVIDIA Remote Attestation Service (NRAS) for verification, and returns an AttestationResult that includes an overall boolean result, an NRAS JWT, and per-device tokens. The package exposes a generic RemoteAttester[T DeviceInfo] that works for both GPUs and switches, GPU- and switch-admins for evidence collection and lifecycle (enable ready state, persistence mode checks, shutdown), and an NRAS client for remote attestation and JWT verification. Example usage shows nonce generation, creating admins and a RemoteAttester, and calling Attest(ctx, nonce). For the AI/ML community this simplifies integrating hardware-rooted trust into deployment pipelines and confidential-compute workflows—useful for multi-tenant inference, model/IP protection, and regulatory compliance. Technically, it ties on-device attestation (NVML/NSCQ) to cloud verification (NRAS) and exposes Go-friendly primitives for automation and CI (unit and hardware-tagged integration tests). By producing verifiable JWTs and per-device tokens, go-nvtrust enables orchestration systems and runtime services to gate workloads on cryptographic proof that GPUs/NVSwitches are in a trusted, confidential-compute state.