Test
Nebius AI Cloud Engineer
Crea una cuenta gratis para terminar este test
Guarda tu progreso y consulta tu resultado final.
¿Ya tienes cuenta? Inicia sesión
Preguntas del test (38)
1A team creates a GPU VM, but the required GPU type does not appear as available. The VM configuration otherwise looks correct. What should you check first?
- AWhether the service account key has expired
- BWhether the correct project and region are selected
- CWhether the VM has encryption at rest enabled
- DWhether the user has the Owner role
2A developer needs permission to create and manage resources in a project but must not be able to grant access to other users. Which role is appropriate?
- AViewer
- BEditor
- COwner
- DAuditor
3A nightly training workflow needs to read data from Object Storage. What is the correct authentication and authorization pattern?
- AUse a developer’s personal account and store the credentials in the image
- BMake the bucket public so the job does not need credentials
- CUse a service account that belongs to a group with the required permissions
- DAssign permissions directly to the workload without using groups
4A project was accidentally created in the wrong region. It is still empty. What is the correct action?
- AChange the project’s region in the console
- BMove the project into another tenant
- CDelete the project and recreate it in the correct region
- DCreate a second region inside the same project
5A development GPU VM is only needed from 9 AM to 5 PM on weekdays. The team wants to reduce cost while keeping the VM available for reuse. What should they do?
- ADelete the VM every evening and recreate it every morning
- BKeep it running because GPU VMs cannot be paused
- CStop it outside working hours and start it again when needed
- DMove it to a different project overnight
6A newly created GPU VM is running, but the engineer wants to confirm that the GPU and NVIDIA driver are functioning correctly. What should they run first?
- Aibstat
- Bnvidia-smi
- Ckubectl describe node
- Dall_reduce_perf
7Two VMs in a distributed training workload need to access the same training files at the same time. Which storage type is the best fit?
- ALocal NVMe
- BNetwork SSD disk
- CShared filesystem
- DObject Storage lifecycle archive
8A Kubernetes node group uses InfiniBand, and you are manually enabling NVIDIA GPU support rather than using the Nebius GPU-driver image. Which installation order is correct?
- AGPU Operator first, then Network Operator
- BNetwork Operator first, then GPU Operator
- CDevice plugin only; neither operator is required
- DGPU Operator only because it includes InfiniBand support
9A team needs a public IP address that must remain reserved even if the VM using it is deleted. What should they use?
- AA dynamic public IP
- BA static route
- CA public IP allocation
- DA security-group reservation
10A multi-node NCCL training job hangs during collective communication. InfiniBand links appear up and basic connectivity has already been validated. What is the most useful next diagnostic action?
- ASet NCCL_DEBUG=INFO
- BRecreate the entire project in another region
- CReplace the VM boot disk
- DChange the Docker subnet
11How do users and service accounts receive permissions in Nebius AI Cloud?
- ABy attaching roles directly to each user or service account
- BThrough membership in groups that hold access permits
- CThrough project membership that automatically grants Editor access
- DBy creating authentication keys tied directly to project roles
12When you grant a group access to a project, what does the access permit bind together?
- AA user identity and a project
- BA group and a service account key
- CA resource, such as a project, and a role
- DA project and an authentication method
13A teammate must review resources in a project but must not modify anything. Which role fits?
- AEditor
- BViewer
- COwner
- DRead-Only Operator
14What can the Owner role do that Editor cannot?
- ACreate and manage project resources
- BRead project resources
- CManage access for other users
- DAuthenticate applications using service-account keys
15A nightly training job must read data from Object Storage without human interaction. Which identity should it use?
- AA personal user account placed in the Viewer group
- BA shared team account with an access permit
- CA service account added to a group with the required permissions
- DA service account with permissions assigned directly to the bucket
16How does an application authenticate as a service account?
- AWith the user credentials of the account that created it
- BWith a project-level access permit
- CWith keys created in the service account’s Keys section
- DWith the group ID of the group containing the service account
17Under the shared-responsibility model, which task belongs to the customer?
- AReplacing failed physical GPU hardware
- BMaintaining datacenter facilities and physical hosts
- CConfiguring IAM access and protecting project data and credentials
- DMaintaining the underlying Nebius platform layer
18Which practice best protects credentials used by automated workloads?
- AStore them in the application image with restricted permissions
- BCommit them to a private repository accessible only to the team
- CStore them in a secret-management mechanism and inject them at runtime
- DStore them in the service account description so they stay associated with the identity
19Why enable encryption at rest for disks and buckets?
- AIt prevents unauthorized users from receiving IAM permissions
- BIt removes the need to restrict access to the stored data
- CIt protects stored data if the underlying media or storage is accessed without authorization
- DIt encrypts all application traffic between VMs automatically
20A development GPU VM is used only during working hours but currently runs 24/7. Which action most directly reduces cost while preserving the VM for continued use?
- ADelete the VM each evening and recreate it each morning
- BMove the VM into a lower-cost project outside working hours
- CStop the VM outside working hours and start it again when needed
- DDetach its GPU while keeping the VM running
21Where in the Nebius console do you review billing and audit logs?
- AProject Settings
- BMonitoring
- CManage
- DIAM and Access
22You created a VM but the GPU type you need is not available. What is the most likely reason?
- AThe VM preset does not support GPU attachment in the selected project
- BThe wrong region is selected, and GPU types or capacity differ by region
- CThe project does not have a default network configured
- DThe VM was created without a GPU-enabled boot disk
23Every resource in Nebius AI Cloud is scoped by which combination?
- ARegion → Tenant → Project → Resources
- BTenant → Project → Resources
- CProject → Tenant → Resources
- DTenant → Resources → Project
24You are about to create a GPU training resource in a shared environment where a colleague is running separate experiments. What should you verify first?
- AThat the VM image contains the latest NVIDIA driver
- BThat the correct security group is attached
- CThat the correct project and region are selected
- DThat the project Owner is currently logged in
25After connecting to a newly created GPU VM over SSH, which command should you run first to confirm the GPU is visible and the NVIDIA driver responds?
- Aibstat
- Bkubectl describe node
- Cnvidia-smi
- Dnvidia-smi topo -m
26You forgot to configure SSH access when creating a VM. What is the primary effect?
- AThe VM fails to boot because no user credentials were provided
- BThe VM still runs, but you cannot easily connect to it from your local machine
- CThe GPU driver cannot initialize until SSH access is added
- DThe VM loses access to its project network
27Training checkpoints are stored on a disk that must be detached from VM A and attached to VM B. Which storage type is the best fit?
- ALocal NVMe
- BNetwork SSD disk
- CShared filesystem
- DObject Storage
28Which statement about local NVMe storage is correct?
- AIt is network-attached storage that can be moved between VMs
- BIt is shared by multiple VMs for concurrent access
- CIt is physically attached to one host and provides very high performance
- DIt is intended mainly for long-term durable datasets
29Several VMs in a distributed training setup need to access the same files concurrently. Which storage type is the best fit?
- ANetwork SSD disk
- BLocal NVMe
- CShared filesystem
- DObject Storage archive class
30How is data in Object Storage typically accessed?
- AAs mounted block devices attached directly to one VM
- BAs shared POSIX files mounted simultaneously across nodes
- CAs complete objects uploaded and downloaded over the network
- DAs host-local files cached on NVMe
31Which Object Storage feature can automatically move objects between storage classes over time?
- AReplication rules
- BBucket versioning
- CLifecycle policies
- DAccess permits
32For pods to consume GPUs on a Kubernetes cluster, what must be present on the GPU nodes?
- AOnly the NVIDIA driver, because Kubernetes detects GPUs automatically
- BOnly the device plugin, because the driver is provided by the pod
- CNVIDIA driver enablement plus a device plugin that exposes GPU resources to the scheduler
- DA shared filesystem plus an InfiniBand network interface
33Which pod specification requests exactly one GPU?
- Aresources.requests: nvidia.com/gpu: 1 only
- Bdevices.limits: gpu: 1
- Cresources.limits with nvidia.com/gpu: 1
- DnodeSelector: nvidia.com/gpu: 1
34After adding a GPU node group to a Kubernetes cluster, how do you confirm that GPUs are schedulable?
- Anvidia-smi shows the GPU model on your local workstation
- Bkubectl get pods shows the node in Running state
- Ckubectl describe node shows nvidia.com/gpu under allocatable capacity
- Dibstat shows the InfiniBand port as Active
35Why are GPU cluster nodes interconnected with InfiniBand?
- ATo provide persistent storage replication between nodes
- BTo provide Kubernetes control-plane connectivity
- CTo provide low-latency, high-bandwidth communication for distributed training traffic such as NCCL collectives
- DTo allow Docker images to be pulled directly from Container Registry
36Which is a standard end-to-end way to validate interconnect performance across GPU nodes?
- ARun nvidia-smi on each node independently
- BRun ibstat and compare port states
- CRun ib_write_bw only between two hosts
- DRun NCCL tests such as all_reduce_perf across the nodes
37Which command order is required to publish a local Docker image to Container Registry?
- Adocker push first, then docker tag with the registry address
- Bdocker tag with the registry address, then docker push
- Cdocker login, then docker run, then docker tag
- Ddocker build, then docker pull, then docker push
38You push a new image using a tag that already exists in the registry. What happens?
- AThe push is rejected because registry tags are immutable
- BThe older image version is deleted and replaced permanently
- CThe tag is updated to point to the newer image version, while earlier versions remain available
- DA second tag is created automatically with a version suffix
Información
- Descripción
- Test sobre Nebius AI Cloud Engineer, creado por Muhammad J Khan el 17/09/2026.
- Autor
- Muhammad J Khan
- Idioma
- English
- Visitas
- 0
- Etiquetas
- Valoración
- Aún sin valoraciones Valora este test