Attesting the GPU

The Virtual Machine User runs confidential applications in a CVM that is configured for Confidential Computing. The Virtual Machine User might also be the Virtual Machine Administrator. NVIDIA recommends that you work in the /shared directory in the guest.

Before you run workloads, attest the GPU to confirm that the system is authentic and valid. During attestation, the GPU collects and signs measurements, and the verifier compares the measurements to known-good reference measurements. After attestation succeeds, set the GPU ready state.

The GPU doesn’t accept work until a user in the CVM sets the ready state. This restriction prevents use of the GPU before you confirm that the GPU is trustworthy. A succcessful attestation does not set the ready state automatically. After attestation succeeds, set the ready state with the following command:

$ nvidia-smi conf-compute -srs 1

Validating the Configuration

If you didn’t configure the guest yourself, confirm that nvidia-persistenced is running:
```
$ ps -aux | grep nvidia-persistenced
root        2327 20.1  0.0   5312  1788 ?        Ss   08:57   0:05 nvidia-persistenced
```
If the output shows only the grep command, start nvidia-persistenced:
```
$ sudo nvidia-persistenced
```
Confirm that the GPU runs in the Confidential Computing mode that you expect:
```
$ nvidia-smi conf-compute -f
CC status: ON
```

Install the NVIDIA Attestestion SDK (NVAT) and attest the GPU. Refer to the NVAT documentation for installation and attestation instructions.
After attestation succeeds, enable the GPU to accept work:
```
$ nvidia-smi conf-compute -srs 1
```