I85PREP
⌘ K
Nutanix NCP-AI study materials
Nutanix NCP-AI study resources

Nutanix NCP-AI practice questions

One question at a time. Build your Nutanix NCP-AI understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Nutanix AI Stack Architecture and Components

In the Nutanix AI stack, which component is responsible for providing the distributed storage fabric that backs persistent volumes used by GPU worker nodes?

  1. Nutanix Flow Network Security
  2. Nutanix AOS with the Distributed Storage Fabric Correct answer.
  3. NVIDIA GPU Operator
  4. Nutanix Prism Central only

Explanation

AOS runs on every node and pools local disks into the Distributed Storage Fabric, which presents datastore and volume services to workloads including Kubernetes persistent volumes. Prism Central is a management plane, Flow is security, and the GPU Operator manages device drivers, not storage.

Question 2Nutanix AI Stack Architecture and Components

Which Nutanix product provides a Kubernetes distribution that is natively integrated with AOS storage classes and Prism-based lifecycle management for AI workloads?

  1. Nutanix NKP (Nutanix Kubernetes Platform) Correct answer.
  2. Nutanix Files
  3. Nutanix Mine
  4. Nutanix Era

Explanation

NKP is Nutanix's Kubernetes platform, integrated with Prism for cluster lifecycle and with AOS-backed CSI storage classes. Files is scale-out NAS, Era is database automation, and Mine is backup target integration.

Question 3Nutanix AI Stack Architecture and Components

A team wants Nutanix-managed GPU virtualization so that a single physical GPU can be partitioned across multiple AI inference pods with memory isolation. Which capability should be enabled on the AHV cluster?

  1. NVIDIA vGPU with mediated device or vGPU profiles on AHV Correct answer.
  2. CPU pinning in the VM custom specification
  3. AHV network segmentation with VLANs
  4. Volume group flash mode

Explanation

AHV supports NVIDIA vGPU, allowing a physical GPU to be time-sliced or partitioned into profiles assigned to multiple VMs and, through passthrough to Kubernetes, to pods. CPU pinning, VLANs, and flash mode affect compute scheduling, networking, and storage performance respectively, not GPU partitioning.

Question 4Nutanix AI Stack Architecture and Components

Which Nutanix component supplies the hyperconverged compute layer on which the AI stack's GPU-bearing VMs and NKP worker nodes run?

  1. Nutanix Objects
  2. Nutanix Calm
  3. Prism Element
  4. AHV Correct answer.

Explanation

AHV is the native Nutanix hypervisor that schedules and runs VMs, including GPU-enabled nodes for AI workloads. Prism Element is the per-cluster management UI, Objects is S3-compatible object storage, and Calm is application automation.

Question 5Nutanix AI Stack Architecture and Components

An architect needs S3-compatible object storage to hold large training datasets that NKP workloads access through a CSI driver. Which Nutanix service fits this requirement?

  1. Nutanix Volumes
  2. Nutanix Objects Correct answer.
  3. Nutanix Flow
  4. Nutanix Leap

Explanation

Nutanix Objects exposes an S3-compatible API and can back buckets used by data pipelines and CSI drivers for object storage. Volumes provides block storage, Flow is network security, and Leap handles disaster recovery.

Question 6Nutanix AI Stack Architecture and Components

Which Prism Central capability lets an administrator define a reusable AI environment template that provisions NKP clusters plus GPU VMs across multiple Nutanix clusters?

  1. AOS capacity runway reporting
  2. Nutanix Calm blueprints driven from Prism Central Correct answer.
  3. Nutanix Flow policy hit counters
  4. Prism Central analysis dashboards

Explanation

Calm blueprints in Prism Central model multi-tier applications, including Kubernetes clusters and GPU VMs, and can deploy them across registered clusters. Dashboards, flow hit counters, and capacity runway are observability features, not provisioning automation.

Question 7Nutanix AI Stack Architecture and Components

What is the primary role of Prism Central in a multi-cluster Nutanix AI deployment?

  1. Replacing the hypervisor on each node
  2. Centralized management, monitoring, and policy across registered clusters Correct answer.
  3. Executing GPU compute for training jobs
  4. Serving the S3 API for dataset buckets

Explanation

Prism Central is the multi-cluster management plane that aggregates inventory, health, and policies such as categories and Flow rules across registered Prism Elements. It does not run GPU jobs, serve object APIs, or replace AHV.

Question 8Nutanix AI Stack Architecture and Components

Which Nutanix feature should be used to isolate GPU training traffic from inference traffic between namespaces on NKP?

  1. Prism Central role-based access control
  2. Nutanix Flow network policies with category-based rules Correct answer.
  3. Nutanix Files share permissions
  4. AOS erasure coding settings

Explanation

Flow applies distributed firewall policy to traffic flows and can target workloads by category, which maps well to separating training and inference namespaces. Erasure coding affects storage efficiency, RBAC controls user permissions, and Files permissions govern SMB/NFS access.

Question 9Nutanix AI Stack Architecture and Components

Which statement best describes how the Nutanix AI stack typically consumes GPU resources on AHV?

  1. GPUs are pooled into the Distributed Storage Fabric and consumed as block devices
  2. Physical GPUs are attached to VMs via passthrough or partitioned with vGPU profiles, and Kubernetes schedules pods onto those nodes Correct answer.
  3. GPUs are exposed only through Nutanix Objects buckets
  4. GPUs are virtualized exclusively by Prism Central firmware

Explanation

AHV attaches GPUs to VMs either by PCI passthrough or by NVIDIA vGPU profiles, and the Kubernetes device plugin then advertises those resources for pod scheduling. GPUs are compute devices and are never part of the storage fabric or object service.

Question 10Nutanix AI Stack Architecture and Components

A customer needs to run NKP with persistent volumes for model checkpoints and requires the storage to survive node failure. Which Nutanix design satisfies this?

  1. NKP with hostPath volumes on each worker node
  2. NKP with NFS exported from a single worker node
  3. NKP with emptyDir volumes sized to GPU memory
  4. NKP with the Nutanix CSI driver using AOS-backed storage classes with replication factor Correct answer.

Explanation

The Nutanix CSI driver provisions persistent volumes from AOS, and the container's replication factor ensures data redundancy across nodes so checkpoints survive a node failure. hostPath and single-node NFS create single points of failure, and emptyDir is ephemeral.

Question 11Nutanix AI Stack Architecture and Components

Which Nutanix capability provides GPU-aware scheduling visibility so administrators can see which NKP nodes have unallocated GPU resources in Prism Central?

  1. Nutanix Era snapshot schedules
  2. AOS erasure coding status
  3. Nutanix Mine backup reports
  4. Prism Central integrated Kubernetes and GPU inventory views Correct answer.

Explanation

Prism Central surfaces registered NKP clusters and node resource inventory, including GPU allocation, so administrators can identify nodes with free accelerators. Mine, Era, and erasure coding status report on backup, database copies, and storage redundancy respectively.

Question 12Nutanix AI Stack Architecture and Components

Which Nutanix product is designed to provide file services for shared datasets accessed by both GPU VMs and NKP pods over NFS or SMB?

  1. Nutanix Objects
  2. Nutanix Files Correct answer.
  3. Nutanix Flow
  4. Nutanix Calm

Explanation

Nutanix Files is the scale-out file services product exposing NFS and SMB shares suitable for shared datasets. Objects is S3-compatible, Flow is security, and Calm is automation.

Question 13Nutanix AI Stack Architecture and Components

What does the Nutanix GPU Operator integration on NKP primarily automate?

  1. Provisioning of AOS storage containers
  2. Deployment of NVIDIA drivers, container toolkit, and device plugin on GPU nodes Correct answer.
  3. Configuration of S3 bucket lifecycle policies
  4. Creation of Prism Central categories

Explanation

The GPU Operator manages the lifecycle of NVIDIA software components on Kubernetes nodes, including drivers, the container runtime toolkit, and the device plugin that advertises GPUs. It does not manage AOS storage containers, Prism categories, or object bucket policies.

Question 14Nutanix AI Stack Architecture and Components

An AI team must guarantee that training pods never share a GPU with inference pods on the same node. Which Kubernetes mechanism enforces this on NKP?

  1. AOS replication factor settings
  2. Prism Central role mapping
  3. Nutanix Flow application rules
  4. Resource requests and limits for nvidia.com/gpu plus node affinity or taints and tolerations Correct answer.

Explanation

Kubernetes schedules GPUs as extended resources; requesting whole GPUs plus taints, tolerations, or node affinity keeps training and inference pods on separate nodes. Flow, replication factor, and role mapping do not influence GPU allocation.

Question 15Nutanix AI Stack Architecture and Components

Which Nutanix component provides the management API and UI used to register multiple clusters and apply global policies for an AI platform spanning data centers?

  1. AHV Turbo
  2. Prism Element
  3. Nutanix Volumes
  4. Prism Central Correct answer.

Explanation

Prism Central registers multiple Prism Element clusters and provides global policy, categories, and RBAC across them. Prism Element manages a single cluster, AHV Turbo is a storage acceleration mode, and Volumes provides block storage.

Question 16Nutanix AI Stack Architecture and Components

Which design choice best supports low-latency model training on the Nutanix AI stack when datasets are read repeatedly by many GPU nodes?

  1. Place data on AOS-backed persistent volumes with data locality to the consuming nodes Correct answer.
  2. Disable the CSI driver and use emptyDir volumes
  3. Store data on a single NFS server in a remote site
  4. Use erasure coding with the maximum parity setting for all volumes

Explanation

AOS keeps data local to the node running the workload where possible, reducing network hops for repeated reads. Remote single-server NFS adds latency and a bottleneck, maximum parity increases write overhead, and emptyDir discards data.

Question 17Nutanix AI Stack Architecture and Components

Which Nutanix capability should be used to recover the NKP control plane and GPU worker VMs after a site outage?

  1. Nutanix Leap with protection policies and recovery plans Correct answer.
  2. Nutanix Calm with blueprints only
  3. Prism Central analysis charts
  4. Nutanix Flow with application rules

Explanation

Leap orchestrates replication and recovery plans for VMs, which can include NKP control plane and worker nodes. Flow is security, Calm alone does not provide DR replication, and analysis charts are reporting only.

Question 18Nutanix AI Stack Architecture and Components

Which statement accurately describes the relationship between Prism Element and AHV on a Nutanix cluster?

  1. Prism Element is the hypervisor and AHV is its management UI
  2. Prism Element and AHV are both object storage services
  3. AHV runs only inside Prism Central VMs
  4. Prism Element is the cluster management service that configures and monitors the AHV hypervisor and its VMs Correct answer.

Explanation

Prism Element is the per-cluster management service, while AHV is the hypervisor that executes VMs; Prism Element configures and monitors AHV. The other options invert or misclassify these roles.

Question 19Nutanix AI Stack Architecture and Components

A GPU node in an NKP cluster reports that pods are pending with an insufficient nvidia.com/gpu error. Which Nutanix-layer check is most relevant first?

  1. Increase the AOS replication factor on the container
  2. Verify the VM has a GPU attached via passthrough or vGPU profile and that the device plugin is healthy Correct answer.
  3. Raise the Prism Central role permissions for the namespace
  4. Add a Flow rule allowing port 443

Explanation

Pending pods requesting nvidia.com/gpu usually mean the node advertises no GPU resource, which traces back to the VM lacking a GPU device or the device plugin not running. Replication factor, Flow rules, and RBAC do not affect advertised GPU capacity.

Question 20Nutanix AI Stack Architecture and Components

Which Nutanix service provides block storage to VMs and can be used for high-throughput scratch space on GPU nodes?

  1. Nutanix Flow
  2. Nutanix Volumes Correct answer.
  3. Nutanix Objects
  4. Nutanix Files

Explanation

Nutanix Volumes presents block storage, typically as iSCSI targets, suitable for scratch or database workloads needing raw block devices. Objects is S3, Files is NFS/SMB, and Flow is security.

Question 21Nutanix AI Stack Architecture and Components

How does Nutanix Calm contribute to an AI platform built on NKP?

  1. It replaces the Kubernetes scheduler
  2. It stores training datasets in S3 buckets
  3. It provides the GPU driver container images
  4. It models multi-tier applications as blueprints and automates their deployment and lifecycle across clusters Correct answer.

Explanation

Calm blueprints describe application tiers, dependencies, and actions, enabling repeatable deployment and lifecycle operations across registered clusters. It does not ship GPU drivers, replace the scheduler, or provide object storage.

Question 22Nutanix AI Stack Architecture and Components

Which Nutanix feature allows an administrator to tag GPU VMs and NKP namespaces with categories so that policy and automation can target them consistently?

  1. AHV Turbo mode
  2. Prism Central categories applied across clusters Correct answer.
  3. Nutanix Volumes CHAP secrets
  4. AOS erasure coding profiles

Explanation

Prism Central categories are key-value labels that span registered clusters and are consumed by Flow, Calm, and other policy engines for consistent targeting. Erasure coding, CHAP secrets, and Turbo mode relate to storage behavior and authentication.

Question 23Nutanix AI Stack Architecture and Components

An organization wants a single pane of glass for capacity planning across several Nutanix clusters hosting AI workloads. Which Prism Central capability supports this?

  1. Capacity and performance analysis with runway forecasting Correct answer.
  2. Flow policy simulation only
  3. Nutanix Mine job scheduling
  4. AHV live migration logs

Explanation

Prism Central analysis aggregates capacity and performance metrics across clusters and forecasts runway so planners can anticipate exhaustion. Flow simulation covers policy, Mine covers backup, and live migration logs are operational events.

Question 24Nutanix AI Stack Architecture and Components

Which statement correctly distinguishes Nutanix Objects from Nutanix Files in an AI data pipeline?

  1. Both serve only block storage over iSCSI
  2. Objects serves an S3-compatible API, while Files serves NFS and SMB shares Correct answer.
  3. Objects serves NFS, while Files serves S3
  4. Both are network security products

Explanation

Objects is the S3-compatible object store for bucket-based pipelines, while Files provides traditional file protocols NFS and SMB. The other options swap or misclassify the protocols.

Question 25Nutanix AI Stack Architecture and Components

Which Nutanix component enforces microsegmentation policy between the NKP control plane and GPU worker nodes without requiring agent installation inside each pod?

  1. Nutanix Calm blueprint audits
  2. AOS replication factor
  3. Nutanix Flow distributed firewall at the AHV layer Correct answer.
  4. Nutanix Files share ACLs

Explanation

Flow enforces policy in the AHV data path, so traffic between VMs hosting Kubernetes nodes is filtered without per-pod agents. File ACLs, Calm audits, and replication factor do not enforce network policy.

Question 26Nutanix AI Stack Architecture and Components

In the Nutanix AI stack, which component is the Kubernetes distribution that Nutanix supports for running GPU-accelerated container workloads on AHV?

  1. Nutanix Objects
  2. Nutanix Prism Element
  3. Nutanix Kubernetes Platform (NKP) Correct answer.
  4. Nutanix Flow Network Security

Explanation

NKP is the Nutanix-supported Kubernetes distribution used to run containerized AI workloads on the Nutanix platform, including GPU-accelerated pods scheduled onto worker VMs running on AHV. The other options are infrastructure services, not Kubernetes distributions.

Question 27Nutanix AI Stack Architecture and Components

Which Nutanix hypervisor hosts the worker virtual machines that make up an NKP cluster in a standard Nutanix AI deployment?

  1. Nutanix Files
  2. Nutanix Calm
  3. Nutanix AHV Correct answer.
  4. Nutanix Era

Explanation

NKP worker and control plane nodes run as VMs on AHV, the native Nutanix hypervisor, which passes through or virtualizes GPUs to those VMs. Files, Calm, and Era are data services and automation products rather than hypervisors.

Question 28Nutanix AI Stack Architecture and Components

A GPU-enabled VM on AHV must expose its physical GPU directly to an NKP worker so that the NVIDIA device plugin can advertise nvidia.com/gpu to the scheduler. Which configuration on the VM enables this?

  1. Enabling deduplication on the VM's storage container
  2. Adding the VM to a protection domain
  3. Attaching a GPU device with passthrough or vGPU configuration to the VM Correct answer.
  4. Assigning the VM a static IP address on the management network

Explanation

For Kubernetes to see a GPU, the VM must actually own a GPU device, which on AHV means configuring GPU passthrough or NVIDIA vGPU on that VM. Storage, networking, and protection domain settings do not present GPU resources to the guest operating system.

Question 29Nutanix AI Stack Architecture and Components

Which Nutanix storage service is the natural target for an S3-compatible bucket holding training datasets and model artifacts consumed by NKP workloads?

  1. Nutanix Volumes
  2. Nutanix Flow
  3. Nutanix Mine
  4. Nutanix Objects Correct answer.

Explanation

Nutanix Objects provides an S3-compatible object store, which is the standard interface for dataset and model artifact buckets used by AI pipelines. Volumes, Flow, and Mine address block access, networking, and backup integration respectively.

Question 30Nutanix AI Stack Architecture and Components

An administrator wants NKP pods to mount a shared read-write dataset that must be accessible concurrently from GPU VMs and from pods on multiple worker nodes. Which Nutanix storage integration best meets this requirement?

  1. A local-path persistent volume bound to a single worker node
  2. A vSphere-only datastore driver
  3. The Nutanix Files CSI driver providing ReadWriteMany persistent volumes Correct answer.
  4. An emptyDir volume mounted into each pod

Explanation

ReadWriteMany access across nodes requires a shared file system, and the Nutanix Files CSI driver provides exactly that for Kubernetes persistent volumes. Local-path, emptyDir, and a vSphere-only driver cannot deliver multi-node shared read-write access on AHV.

Question 31Nutanix AI Stack Architecture and Components

Which statement best describes the role of Prism Central in the Nutanix AI stack architecture?

  1. It provides centralized multi-cluster management, categories, and policy that span the clusters hosting NKP and GPU workloads Correct answer.
  2. It provides the NVIDIA device plugin that advertises GPUs to Kubernetes
  3. It serves as the container runtime for NKP worker nodes
  4. It replaces AHV as the runtime for GPU virtual machines

Explanation

Prism Central is the centralized management plane across registered Prism Element clusters, enabling categories, global policies, and cross-cluster visibility for AI environments. It does not run workloads, provide a container runtime, or implement the NVIDIA device plugin.

Question 32Nutanix AI Stack Architecture and Components

Which Nutanix capability allows an administrator to apply consistent configuration and policy to GPU VMs across multiple clusters by labeling them with key-value pairs?

  1. Categories in Prism Central Correct answer.
  2. Alert policies in Prism Central
  3. Storage containers in Prism Element
  4. Protection domains in Prism Element

Explanation

Categories are key-value labels assigned to VMs and other entities in Prism Central and are used as the selector for policies, playbooks, and automation across clusters. Protection domains, storage containers, and alert policies serve replication, storage layout, and monitoring purposes.

Question 33Nutanix AI Stack Architecture and Components

A team runs both latency-sensitive inference and long batch training on the same NKP cluster. Which architectural approach best isolates GPU capacity between the two workload classes?

  1. Dedicated node pools with taints and tolerations so each workload class schedules only onto its own GPU nodes Correct answer.
  2. Running all workloads as privileged pods on the control plane nodes
  3. Disabling the NVIDIA device plugin so GPUs are not advertised
  4. A single shared node pool with no taints so the scheduler can pack all pods together

Explanation

Separate node pools combined with taints and tolerations keep each workload class on its own GPU nodes, preventing batch training from consuming capacity reserved for inference. Sharing one untainted pool, hiding GPUs, or running workloads on control plane nodes undermines isolation and stability.

Question 34Nutanix AI Stack Architecture and Components

Which Nutanix feature provides distributed firewall enforcement between NKP namespaces without requiring changes to the physical network?

  1. Nutanix Objects
  2. Nutanix Calm
  3. Nutanix Files
  4. Nutanix Flow Network Security Correct answer.

Explanation

Flow Network Security applies distributed firewall policy at the virtual NIC level on AHV, allowing namespace and VM-level isolation without reconfiguring physical switches. Files and Objects are storage services, and Calm is an automation product.

Question 35Nutanix AI Stack Architecture and Components

An AI platform team wants repeatable, self-service deployment of an NKP cluster plus its GPU worker VMs and supporting storage, driven from a blueprint. Which Nutanix component fits this role?

  1. Nutanix Objects
  2. Nutanix Flow
  3. Nutanix Calm Correct answer.
  4. Nutanix Prism Element

Explanation

Calm provides blueprint-based application automation so that multi-tier environments such as an NKP cluster with GPU workers and storage can be deployed repeatably and self-service. Prism Element, Objects, and Flow provide cluster management, object storage, and network security rather than blueprint orchestration.

Question 36Nutanix AI Stack Architecture and Components

Which statement correctly describes how Nutanix Move or Nutanix Cloud Clusters (NC2) relate to the AI stack architecture?

  1. They provide the NVIDIA GPU driver stack inside NKP worker nodes
  2. They replace Prism Central as the multi-cluster management plane
  3. They supply the CSI driver used for persistent volumes
  4. They extend or migrate workloads onto Nutanix infrastructure so that NKP and GPU VMs can run on AHV in the target environment Correct answer.

Explanation

Move and NC2 address workload migration and hybrid cloud extension onto Nutanix infrastructure, which is where NKP and GPU VMs then run on AHV. They do not deliver GPU drivers, replace Prism Central, or implement storage CSI drivers.

Question 37Nutanix AI Stack Architecture and Components

Which Nutanix capability should an administrator use to confirm that a cluster has enough free GPU capacity before onboarding a new NKP training project?

  1. Reviewing the Flow security policy hit counts
  2. Checking the SMB share permissions on Nutanix Files
  3. Reviewing GPU allocation and utilization reporting available through Prism Central Correct answer.
  4. Checking the Objects bucket lifecycle rules

Explanation

Prism Central aggregates capacity and utilization reporting across registered clusters, including GPU consumption, which supports planning before a new project is onboarded. File permissions, firewall hit counts, and bucket lifecycle rules do not report GPU capacity.

Question 38Nutanix AI Stack Architecture and Components

In the Nutanix AI stack, which layer is responsible for installing and managing the NVIDIA driver, container toolkit, and device plugin on NKP worker nodes?

  1. The GPU operator deployed into the Kubernetes cluster Correct answer.
  2. The Nutanix Files CSI driver
  3. Nutanix Objects
  4. Prism Element on the AHV cluster

Explanation

The GPU operator runs inside Kubernetes and automates driver, container toolkit, and device plugin lifecycle on worker nodes so that GPUs become schedulable resources. Prism Element, Objects, and the Files CSI driver operate at the infrastructure and storage layers instead.

Question 39Nutanix AI Stack Architecture and Components

A customer requires that model checkpoints written by NKP pods remain available after the loss of a single AHV node. Which storage design satisfies this?

  1. Persistent volumes backed by a local-path provisioner on each worker
  2. emptyDir volumes shared between pods on the same node
  3. HostPath volumes pointing at the worker's boot disk
  4. Persistent volumes backed by a replicated Nutanix storage container through a CSI driver Correct answer.

Explanation

Nutanix CSI-backed persistent volumes inherit the replication factor of the underlying storage container, so data survives the loss of one node. Local-path, emptyDir, and hostPath volumes are bound to a single node's local storage and are lost or unavailable if that node fails.

Question 40Nutanix AI Stack Architecture and Components

Which Nutanix component is the per-cluster management interface used to configure storage containers and GPU passthrough settings for the VMs that will host NKP workers?

  1. Nutanix Flow
  2. Nutanix Objects
  3. Prism Element Correct answer.
  4. Prism Central

Explanation

Prism Element is the management interface for an individual AHV cluster, where storage containers and VM hardware settings such as GPU passthrough are configured. Prism Central manages multiple clusters, while Objects and Flow provide storage and security services.

Question 41AI Workload Infrastructure and GPU Virtualization

In a Nutanix AHV cluster, which component is responsible for exposing a physical GPU to a guest VM so that the VM can use it for compute workloads?

  1. The Prism Central licensing service
  2. The Flow Network Security policy engine
  3. The AHV hypervisor's GPU passthrough or vGPU configuration on the host Correct answer.
  4. The distributed storage fabric's extent store

Explanation

GPU access for a VM is arranged at the hypervisor layer, either by passing a whole physical GPU through to a single VM or by partitioning it with NVIDIA vGPU software managed on the AHV host. Prism Central, storage, and network services do not allocate GPU devices to guests.

Question 42AI Workload Infrastructure and GPU Virtualization

Which statement best describes the difference between GPU passthrough and NVIDIA vGPU on AHV?

  1. Passthrough only works with AMD GPUs, while vGPU only works with Intel GPUs
  2. Passthrough requires NVIDIA AI Enterprise licensing, while vGPU is always license-free
  3. Passthrough partitions a GPU into slices, while vGPU dedicates the whole card to a single VM
  4. Passthrough dedicates an entire physical GPU to one VM, while vGPU partitions a physical GPU into multiple virtual GPU instances shared across VMs Correct answer.

Explanation

Passthrough gives one VM exclusive ownership of a whole physical GPU, which maximizes isolation but wastes capacity when the workload is small. vGPU uses NVIDIA virtual GPU software to divide a physical card into multiple mediated instances so several VMs can share it.

Question 43AI Workload Infrastructure and GPU Virtualization

A VM configured with GPU passthrough on AHV is being migrated. What is the key limitation an administrator must plan for?

  1. The VM cannot be live-migrated while it holds a passed-through GPU, so it must be powered off and cold-migrated Correct answer.
  2. The VM can be live-migrated only if Flow Network Security is disabled
  3. The VM can be live-migrated only during a Prism Central upgrade window
  4. The VM can be live-migrated freely because passthrough devices follow the VM automatically

Explanation

A passed-through PCI device is bound to a specific physical host, so the guest cannot be live-migrated with its GPU attached. Administrators must power off the VM and cold-migrate it, or use a vGPU configuration that supports migration.

Question 44AI Workload Infrastructure and GPU Virtualization

Which NVIDIA software layer must be installed and licensed on AHV hosts to provide mediated virtual GPU instances to multiple VMs?

  1. NVIDIA CUDA Toolkit alone
  2. NVIDIA Container Toolkit alone
  3. NVIDIA vGPU software with the appropriate vGPU license Correct answer.
  4. NVIDIA Magnum IO alone

Explanation

Mediated vGPU instances require the NVIDIA vGPU host driver and a valid vGPU license tied to the GPU model and instance type. CUDA, Container Toolkit, and Magnum IO are user-space or communication libraries and do not create virtual GPU devices.

Question 45AI Workload Infrastructure and GPU Virtualization

An administrator wants a single VM to use a fraction of a physical data center GPU for a small inference workload while leaving capacity for other VMs. Which configuration is most appropriate?

  1. Disable GPU access and rely on CPU inference
  2. Attach multiple physical GPUs via passthrough to the same VM
  3. Assign a vGPU profile sized to the workload's memory and compute needs Correct answer.
  4. Assign the entire physical GPU via passthrough

Explanation

A vGPU profile allocates a defined slice of GPU memory and compute to the VM, allowing the remaining capacity to serve other VMs. Passthrough would consume the whole card for one guest, which is wasteful for a small workload.

Question 46AI Workload Infrastructure and GPU Virtualization

What does the frame rate limiter setting control in an NVIDIA vGPU configuration on AHV?

  1. The maximum number of VMs that can share a single physical GPU
  2. The maximum PCIe bandwidth allocated to the vGPU
  3. The maximum display refresh rate delivered to virtual display sessions Correct answer.
  4. The maximum CUDA core clock of the physical GPU

Explanation

The frame rate limiter caps the refresh rate presented to virtual display sessions, which is useful for VDI density tuning. It does not change clock speeds, the number of vGPU instances, or PCIe bandwidth allocation.

Question 47AI Workload Infrastructure and GPU Virtualization

Which Nutanix capability helps an administrator confirm which physical host a GPU-equipped VM is currently running on?

  1. The storage container replication factor
  2. The Flow Network Security policy list
  3. VM host affinity information shown in Prism Element or Prism Central Correct answer.
  4. The cluster's NTP configuration

Explanation

Prism Element and Prism Central expose the host placement of each VM, which lets administrators verify where a GPU-bound VM is running. Storage replication, security policy, and time synchronization settings do not report VM-to-host placement.

Question 48AI Workload Infrastructure and GPU Virtualization

A team is deploying a large language model training job that needs the full memory of an 80 GB GPU and maximum peer-to-peer bandwidth between GPUs on the same host. Which approach is generally preferred?

  1. Sharing a single vGPU profile across many small VMs
  2. Relying on the distributed storage fabric for GPU memory
  3. Passthrough of whole GPUs to the training VM with NVLink-capable GPUs on the same host Correct answer.
  4. Using vGPU time-slicing with the smallest available profile

Explanation

Large training jobs benefit from dedicated whole-GPU memory and high-bandwidth interconnects such as NVLink between GPUs on the same host. Fractional vGPU profiles reduce memory per VM and do not maximize inter-GPU bandwidth.

Question 49AI Workload Infrastructure and GPU Virtualization

Which consideration is most important when planning GPU capacity for an inference service that must serve many small concurrent requests with predictable latency?

  1. Balancing vGPU profile size against the number of concurrent model instances per physical GPU Correct answer.
  2. Minimizing GPU memory per instance regardless of model size
  3. Disabling GPU sharing to guarantee isolation for every request
  4. Maximizing the number of passthrough VMs so each gets a full GPU

Explanation

Inference density depends on fitting enough model instances into GPU memory while preserving latency targets, so profile size must be matched to concurrency. Over-provisioning whole GPUs wastes capacity, while under-sizing memory causes swapping or out-of-memory failures.

Question 50AI Workload Infrastructure and GPU Virtualization

What is the purpose of the NVIDIA GPU Operator when running GPU-accelerated containers on a Kubernetes cluster built on Nutanix infrastructure?

  1. It replaces the AHV hypervisor with a container runtime
  2. It provides the distributed storage fabric for persistent volumes
  3. It manages Prism Central role-based access control
  4. It automates deployment of GPU drivers, device plugins, and related components on cluster nodes Correct answer.

Explanation

The GPU Operator packages and manages the components needed to expose GPUs to Kubernetes workloads, including drivers and the device plugin. It does not replace the hypervisor, provide storage, or handle Prism Central authentication.