Free NVIDIA NCP-AIO Practice Exam with Questions & Answers

Comprehensive and Detailed Explanation From Exact Extract:

NVIDIA Fabric Manager manages the NVLink and NVSwitch fabric resources within HGX systems, enabling efficient resource allocation, communication, and virtualization necessary for AI/ML workloads. This is critical for virtualization as it ensures optimized interconnect performance between GPUs. Video encoding, graphical rendering, or memory upgrades are outside the scope of Fabric Manager.

=============

Comprehensive and Detailed Explanation From Exact Extract:

The NVIDIA Virtual Machine Image (VMI) for machine learning provisioning automatically sets up anUbuntu Serverenvironment with essential components including Docker-CE, NVIDIA Container Toolkit, CSP CLI, NGC CLI, NVIDIA Driver, andRapids—a suite of GPU-accelerated data science and analytics libraries. This comprehensive stack enables immediate development and deployment of ML workloads.

=============

Comprehensive and Detailed Explanation From Exact Extract:

The pre-configured GPU Operator in the NVIDIA Enterprise Catalog differs from the public NGC catalog GPU Operator primarily by its configuration to use aprebuilt vGPU driver imageand being configured to use theNVIDIA License System (NLS). These adaptations allow better support for enterprise environments where vGPU functionality and license management are critical.

Other options such as automatic installation of the Datacenter driver or additional installation of Network Operator are not specific differences highlighted between the two operators.

=============

Comprehensive and Detailed Explanation From Exact Extract:

NVIDIA Fabric Manager is responsible for managing the fabric interconnect in HGX systems, including:

Configuring routing among NVSwitch ports (A)to optimize communication paths.

Coordinating with the NVSwitch driver to train NVSwitch-to-NVSwitch NVLink interconnects (C)for high-speed link setup.

Coordinating with the GPU driver to initialize and train NVSwitch-to-GPU NVLink interconnects (D)ensuring optimal connectivity between GPUs and switches.

Installing the GPU operator and vGPU driver is typically handled separately and not part of Fabric Manager’s core tasks.

=============

Comprehensive and Detailed Explanation From Exact Extract:

To restrict users’ access strictly to the GPUs allocated to their jobs, Slurm usescgroups (control groups)for resource isolation. Enabling device cgroup enforcement by settingConstrainDevices=yesincgroup.confenforces device access restrictions, ensuring jobs cannot access GPUs beyond those assigned.

Increasing memory allocation or setting job priorities does not restrict device access.

Modifying job scripts to request additional CPU cores does not limit GPU access.

Hence, enablingcgroup enforcement with ConstrainDevices=yesis the correct method to prevent users from accessing unallocated GPUs.

Comprehensive and Detailed Explanation From Exact Extract:

Communication issues between UFM server and InfiniBand switches often result from misconfigured or missingsubnet manager configurationon the switches. The subnet manager controls fabric membership and routing, so verifying and correcting its setup is essential for proper UFM operation. Rebooting, adding GPUs, or disabling firewalls are less likely to resolve fabric-level communication problems.

=============

Comprehensive and Detailed Explanation From Exact Extract:

In Kubernetes architecture, thecontrol planeis composed of several core components including thekube-apiserver,etcd(the cluster’s key-value store),kube-scheduler, andkube-controller-manager. These manage the overall cluster state, scheduling, and orchestration of workloads. Theworker nodesare responsible for running the actual containers and include thekubelet(agent that communicates with the control plane) andkube-proxy(handles network routing for services). Other options incorrectly assign these components or roles.

=============

Comprehensive and Detailed Explanation From Exact Extract:

NVIDIA's Multi-Instance GPU (MIG) technology provideshardware-level isolationof critical GPU resources such as memory, cache, and compute units for each GPU instance. This ensures that workloads running in one instance are fully isolated and cannot interfere with the performance of workloads in other instances, supporting multi-tenancy without contention.

=============

Comprehensive and Detailed Explanation From Exact Extract:

To optimize scheduling of multi-GPU jobs in Slurm, it is essential to correctly specify GPU requests in job scripts using--gres=gpu:and enable/configureSlurm’s backfill scheduler. Backfill allows smaller jobs to run opportunistically in gaps without delaying larger multi-GPU jobs, improving cluster utilization and reducing wait times for multi-GPU jobs. Proper configuration ensures efficient packing and priority handling of GPU resources.

=============

Comprehensive and Detailed Explanation From Exact Extract:

For troubleshooting and performance optimization on NVIDIA DGX systems such as DGX H100, theNVIDIA System Management (nvsm)tool is used to gather system health and diagnostic data. The command nvsm dump health is the correct command to generate and export detailed system logs related to the health report of the DGX system.

nvsm show logs --save is not a recognized command format.

nvsm get logs retrieves logs but does not specifically dump the health report logs.

nvsm health --dump-log is not a standard documented nvsm command.

Therefore, nvsm dump health is the valid and documented command used to generate system logs focused on health reporting, useful for diagnosing performance issues in DGX H100 systems.

This usage aligns with NVIDIA’s system management tools guidance for DGX platforms as described in NVIDIA AI Operations documentation for troubleshooting and performance optimization.