Free NVIDIA AI Operations Certification Exam Practice Questions

Sample Practice Questions

Try these 10 sample questions from the NVIDIA AI Operations Certification Exam practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1During a BCM 11 head-node installation, an Ethernet adapter is missing from the Hardware info screen because its required kernel module has not been loaded. What should the administrator do next?

   1. A.Return to Kernel Modules and load the required module.Correct answer: A
   2. B.Assign a static IP before the adapter has a loaded driver.
   3. C.Change the host's DNS name to trigger adapter discovery.
   4. D.Skip the driver and continue using the undetected adapter.

   Explanation: NVIDIA states, "If additional kernel modules are required, then the administrator should go back to the Kernel Modules screen." Loading the module addresses the diagnosed missing driver. IP assignment and hostname changes do not load a hardware driver. Continuing without it does not make the adapter usable.

   Reference: NVIDIA BCM 11 Quickstart Installation Guide — 1.1 Installing The Head Node, Hardware info

2. 2An enterprise must deploy and maintain AI applications across thousands of geographically dispersed edge locations from one console. Which documented role does Fleet Command provide?

   1. A.A centralized cloud control plane for the distributed edge fleet.Correct answer: A
   2. B.A local-only GPU profiler requiring access to each server console.
   3. C.A batch scheduler that makes every edge site one Slurm partition.
   4. D.An NGC image registry that stores images without deploying them.

   Explanation: The official documentation states, "NVIDIA Fleet Command is a centralized control plane that operates in the cloud." Fleet Command centrally manages remote deployments. A local profiler, a batch queue or an image registry alone does not provide the documented fleet-management control plane.

   Reference: NVIDIA Fleet Command — Data Sheet — Physical page 2, Command Your Edge Fleet with One Control Plane

3. 3A Kubernetes Pod requests 6 CPUs, but each eligible node has only 4 CPUs available for scheduling. Which declared value is directly used by the scheduler to assess placement?

   1. A.The container's configured CPU limit.
   2. B.The Pod's CPU resource request.Correct answer: B
   3. C.The node's instantaneous CPU utilization.
   4. D.The Deployment's requested replica count.

   Explanation: The Kubernetes documentation states, "When you specify the resource request for containers in a Pod, the kube-scheduler uses this information to decide which node to place the Pod on." The explicit 6-CPU request cannot fit into the stated 4-CPU scheduling capacity. The other values are not that reservation.

   Reference: Kubernetes — Resource Management for Pods and Containers — Resource Management for Pods and Containers, opening resource-request description

4. 4A GPU container fails during startup. You need NVIDIA Container Runtime debug logs for a reproducible failure. Which change follows NVIDIA's troubleshooting procedure?

   1. A.Set the container's application log level to INFO and inspect only its stdout.
   2. B.Uncomment the debug setting in the runtime config.toml, then reproduce the failure.Correct answer: B
   3. C.Turn on Docker image build progress output, then inspect the cached build layers.
   4. D.Enable GPU persistence mode on the host, then inspect the GPU utilization counters.

   Explanation: NVIDIA instructs, "Edit your runtime configuration under /etc/nvidia-container-runtime/config.toml and uncomment the debug=... line." It then says, "Run your container again to reproduce the issue and generate the logs." These steps target the runtime failure. Application output, image build progress and GPU counters do not enable this runtime diagnostic log.

   Reference: NVIDIA Container Toolkit — Troubleshooting — Generating Debugging Logs

5. 5What happens to existing data on the head-node drive selected at the Disk layout screen of a BCM 11 installation?

   1. A.Its filesystem is mounted without changing existing files.
   2. B.Its data is automatically copied to the cluster's compute nodes.
   3. C.Its previous content is overwritten by the installation.Correct answer: C
   4. D.Its partitions are retained and only missing packages are added.

   Explanation: NVIDIA states, "The installation will be done onto this drive, overwriting all its previous content." Drive selection is destructive installation, not a backup or an in-place package repair. Existing content must not be assumed to survive.

   Reference: NVIDIA BCM 11 Quickstart Installation Guide — 1.1 Installing The Head Node, Disk layout

6. 6An administrator needs temporary remote troubleshooting access to a Fleet Command edge deployment. Which documented security feature governs such access?

   1. A.Application signing that determines how long a remote session remains authorized.
   2. B.Remote-management access controls combined with timed sessions.Correct answer: B
   3. C.Vulnerability scanning that grants a temporary interactive login to the host.
   4. D.Boot measurement that sets each administrator's remote-session duration.

   Explanation: NVIDIA states, "Remote management on Fleet Command offers access controls and timed sessions" Remote authorization and session duration govern this access. Application signing, vulnerability scanning and boot measurement protect different boundaries and do not implement the documented timed remote access.

   Reference: NVIDIA Fleet Command — Advanced Management Features — Advanced Remote Management

7. 7A Linux Kubernetes container repeatedly reaches its CPU limit while remaining otherwise healthy. Which enforcement behavior should the operator expect?

   1. A.The kernel kills it as though it exceeded a memory limit.
   2. B.The scheduler migrates it immediately to a faster node.
   3. C.CPU throttling by the kernel.Correct answer: C
   4. D.The kubelet raises its CPU request to match usage.

   Explanation: The official documentation states, "cpu limits are enforced by CPU throttling." The CPU limit is enforced by throttling; OOM killing is a memory mechanism. Automatic migration or request increases are not this enforcement behavior.

   Reference: Kubernetes — Resource Management for Pods and Containers — Requests and limits, CPU enforcement

8. 8On an SELinux-enabled RHEL host, a documented workaround for an NVML permissions failure is --security-opt=label=disable. What security consequence must an administrator assess before using it?

   1. A.The option disables SELinux separation for that container, which runs unconfined.Correct answer: A
   2. B.The option makes SELinux relabel every mounted host path for exclusive container use.
   3. C.The option preserves SELinux separation while granting access only to GPU counters.
   4. D.The option changes the container's application user to the GPU device's owning user.

   Explanation: NVIDIA warns, "However, using this option disables SELinux separation in the container, and the container runs in an unconfined type." This is a separation tradeoff requiring policy review, rather than a targeted device permission or automatic relabeling/user change.

   Reference: NVIDIA Container Toolkit — Troubleshooting — NVML: Insufficient Permissions and SELinux

9. 9In a standard BCM Type 1 installation, which network does externalnet represent?

   1. A.The private interconnect used only by GPU collectives.
   2. B.The site network, such as the corporate or campus network.Correct answer: B
   3. C.The isolated management network used only for BMC access.
   4. D.The container overlay reserved for Kubernetes pod addresses.

   Explanation: NVIDIA states, "The network externalnet corresponds to the site network that the cluster resides in (for example, a corporate or campus network)." externalnet connects the cluster to its surrounding site. GPU fabrics, BMC networks and pod overlays have different roles and do not match this definition.

   Reference: NVIDIA BCM 11 Quickstart Installation Guide — 1.1 Installing The Head Node, Networks

10. 10At which stage does Fleet Command's documented secure-deployment workflow scan applications for vulnerabilities and malware?

    1. A.After the application has processed its first production dataset.
    2. B.Only after an administrator opens a remote troubleshooting session.
    3. C.Before the applications are loaded.Correct answer: C
    4. D.Only after a device has lost contact with the cloud console.

    Explanation: The official documentation states, "Applications are scanned for vulnerabilities and malware before they’re loaded." Scanning is a pre-load safeguard, not a reaction to first production use, remote access or loss of connectivity.

    Reference: NVIDIA Fleet Command — Data Sheet — Physical page 2, Secure Deployment