Installation and Deployment
Free NVIDIA AI Operations Professional (NCP-AIO) practice — 6 questions on Installation and Deployment, with explanations. No sign-up. Full 12-question mixed test →
A data center team needs to deploy DOCA Services on BlueField-2 DPUs to enable storage acceleration and network security features. The DPUs are installed in host servers, and the team wants to use Base Command Manager to orchestrate the installation. What is the FIRST step required before DOCA Services can be deployed to the DPU Arm cores?
DOCA Services run on the BlueField DPU's Arm cores, which requires the DPU to be configured in separated mode (not embedded mode) where the Arm subsystem operates independently. The Arm cores must have network connectivity to BCM for orchestration and management. This is the foundational prerequisite before any DOCA Services deployment can proceed.
An organization is initializing a new Kubernetes cluster on NVIDIA DGX systems using Base Command Manager. The infrastructure team has completed hardware provisioning and now needs to bootstrap Kubernetes across 12 compute nodes. Which BCM workflow correctly sequences the Kubernetes initialization process?
BCM follows a structured workflow: organize nodes into categories (logical grouping), assign target nodes, provision OS images with required dependencies (container runtime like containerd/Docker), then deploy Kubernetes using BCM templates. The template handles control plane initialization and worker node joining in an orchestrated manner, which is the BCM-native approach for Kubernetes deployment.
A cluster administrator needs to update NVIDIA GPU driver versions across 50 compute nodes to address a critical security vulnerability (CVE-2025-XXXX). The cluster is running mixed workloads with active jobs that cannot be interrupted. Using Base Command Manager, what is the BEST approach to apply this security patch while minimizing workload disruption?
BCM supports rolling update policies that drain nodes (evict or wait for jobs to complete), update a limited batch, then move to the next batch. This approach minimizes disruption by ensuring only a subset of nodes are offline at any time and coordinates with the workload manager to handle job migration or rescheduling. The 5-node batch size prevents overwhelming the remaining cluster capacity.
An AI infrastructure team is installing Run:ai on an existing Kubernetes cluster managed by Base Command Manager. The cluster has 8 DGX A100 nodes with 8 GPUs each. After installing the Run:ai operator and control plane, GPU resources are not appearing in the Run:ai dashboard. Investigation shows that the NVIDIA device plugin is installed but GPUs are not being advertised to Kubernetes. What is the MOST likely root cause and resolution?
For GPU resources to be available in Kubernetes (and thus visible to Run:ai), the container runtime must be configured with nvidia-container-runtime. Without this configuration, containers cannot access GPUs even if the device plugin is installed. The device plugin advertises GPUs to Kubernetes, but the runtime must actually provide GPU access to containers. This is a foundational requirement that must be in place before the device plugin can function properly.
A research organization needs to configure user access controls in Base Command Manager for a new project team of 15 researchers who will submit AI training jobs. The team should be able to view cluster status, submit jobs to specific GPU partitions, and monitor their own jobs, but should NOT be able to modify system configurations or view other teams' job details. Which BCM configuration achieves these requirements?
BCM supports role-based access control (RBAC) where custom roles can be created with specific permissions (job submit, monitor own jobs, view cluster). User groups streamline management for teams, and resource policies allow mapping groups to specific partitions/resources. This approach follows the principle of least privilege and leverages BCM's native RBAC features for fine-grained access control.
An enterprise is deploying Slurm as the workload manager for their NVIDIA AI cluster using Base Command Manager. The Slurm installation must support GPU scheduling with GRES (Generic Resource Scheduling), integrate with BCM for centralized monitoring, and use a dedicated management node separate from the compute nodes. During the installation process using BCM's Slurm deployment template, what configuration is REQUIRED to enable BCM to monitor Slurm job metrics and resource utilization?
BCM integrates with Slurm by querying the Slurm REST API (slurmrestd), which provides access to job information, node states, and resource utilization. The API must be enabled on the slurmctld management node, and BCM must be configured with the endpoint URL and authentication credentials. This is the standard integration method for BCM to monitor Slurm-managed clusters.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$199.99 $34.99 with code FREETEST33 — valid through September 2.
Get my $34.99 deal →