Question # 1

When troubleshooting Slurm job scheduling issues, a common source of problems is jobsgetting stuck in a pending state indefinitely.Which Slurm command can be used to view detailed information about all pending jobs andidentify the cause of the delay?

A. scontrol 
B. sacct 
C. sinfo 

Show Answer

Question # 2

You are managing a Kubernetes cluster running AI training jobs using TensorFlow. Thejobs require access to multiple GPUs across different nodes, but inter-node communicationseems slow, impacting performance.What is a potential networking configuration you would implement to optimize inter-nodecommunication for distributed training?

A. Increase the number of replicas for each job to reduce the load on individual nodes. 
B. Use standard Ethernet networking with jumbo frames enabled to reduce packet overhead during communication.
C. Configure a dedicated storage network to handle data transfer between nodes during training.
D. Use InfiniBand networking between nodes to reduce latency and increase throughput fordistributed training jobs.  

Show Answer

Question # 3

A cloud engineer is looking to provision a virtual machine for machine learning using theNVIDIA Virtual Machine Image (VMI) and Rapids.What technology stack will be set up for the development team automatically when the VMIis deployed?

A. Ubuntu Server, Docker-CE, NVIDIA Container Toolkit, CSP CLI, NGC CLI, NVIDIA Driver
B. Cent OS, Docker-CE, NVIDIA Container Toolkit, CSP CLI, NGC CLI 
C. Ubuntu Server, Docker-CE, NVIDIA Container Toolkit, CSP CLI, NGC CLI, NVIDIA Driver, Rapids
D. Ubuntu Server, Docker-CE, NVIDIA Container Toolkit, CSP CLI, NGC CLI 

Show Answer

Question # 4

You are tasked with deploying a DOCA service on an NVIDIA BlueField DPU in an airgapped data center environment. The DPU has the required BlueField OS version (3.9.0 orhigher) installed, and you have access to the necessary container image from NVIDIA'sNGC catalog. However, you need to ensure that the deployment process is successfulwithout an internet connection.Which of the following steps should you take to deploy the DOCA service on the DPU?

A. Install Docker on the DPU, pull the container directly from NGC, and run it using ‘dockerrun’ with appropriate environment variables.
B. Pull the container image from NGC using Docker and modify the YAML file before deployment.
C. Manually download the container image and YAML file beforehand, transfer them to the DPU, and deploy using Kubernetes with standalone Kubelet.
D. Use the host system’s Docker engine to pull the container image and deploy it on the DPU via SSH. 

Show Answer

Question # 5

You are managing a high-performance computing environment. Users have reportedstorage performance degradation, particularly during peak usage hours when both smallmetadata-intensive operations and large sequential I/O operations are being performedsimultaneously. You suspect that the mixed workload is causing contention on the storagesystem.Which of the following actions is most likely to improve overall storage performance in thismixed workload environment?

A. Reducing stripe count for large files would decrease parallelism, likely worsening performance for large sequential I/O operations.
B. Separate metadata-intensive operations and large sequential I/O operations by using different storage pools for each type of workload.
C. Increase the number of Object Storage Targets (OSTs) to handle more metadata operations.
D. Disable GPUDirect Storage (GDS) during peak hours to reduce I/O load on the Lustre file system.

Show Answer

Question # 6

You are configuring networking for a new AI cluster in your data center. The cluster willhandle large-scale distributed training jobs that require fast communication betweenservers.What type of networking architecture can maximize performance for these AI workloads?

A. Implement a leaf-spine network topology using standard Ethernet switches to ensure scalability as more nodes are added.
B. Prioritize out-of-band management networks over compute networks to ensure efficient job scheduling across nodes.
C. Use standard Ethernet networking with a focus on increasing bandwidth through multiple connections per server.
D. Use InfiniBand networking to provide low-latency, high-throughput communication between servers in the cluster.

Show Answer

Question # 7

What should an administrator check if GPU-to-GPU communication is slow in a distributedsystem using Magnum IO?

A. Limit the number of GPUs used in the system to reduce congestion. 
B. Increase the system's RAM capacity to improve communication speed. 
C. Disable InfiniBand to reduce network complexity. 
D. Verify the configuration of NCCL or NVSHMEM. 

Show Answer

Question # 8

An instance of NVIDIA Fabric Manager service is running on an HGX system with KVM. ASystem Administrator is troubleshooting NVLink partitioning.By default, what is the GPU polling subsystem set to?

A. Every 1 second 
B. Every 30 seconds 
C. Every 60 seconds 
D. Every 10 seconds 

Show Answer

Question # 9

A system administrator is troubleshooting a Docker container that crashes unexpectedlydue to a segmentation fault. They want to generate and analyze core dumps to identify theroot cause of the crash.Why would generating core dumps be a critical step in troubleshooting this issue?

A. Core dumps prevent future crashes by stopping any further execution of the faulty process. 
B. Core dumps provide real-time logs that can be used to monitor ongoing application performance. 
C. Core dumps restore the process to its previous state, often fixing the error-causing crash.
D. Core dumps capture the memory state of the process at the time of the crash. 

Show Answer

Question # 10

An administrator wants to check if the BlueMan service can access the DPU.How can this be done?

A. Via system logs 
B. Via the DOCA Telemetry Service (DTS) 
C. Via a lightweight database operating in the DPU server 
D. Via Linux dump files 

Show Answer