NCP-AI Infrastructure v1.0
Question 1
A cluster administrator needs to validate transceiver firmware versions across 200 ports using UFM.
Which GUI-based method provides a consolidated view?
- A. Use “Topology” view to visually inspect cable icons.
- B. Navigate to “Devices” > select a switch > “Cables” tab to see ASIC firmware and transceiver versions.
- C. Export all switch logs and grep for “FW Version”
- D. Run mlxlink –d lid-<LID> -m on each port manually
Answer : B
Question 2
What command sequence is used to identify the exact name of the server that runs as the master SM in a multi-node fabric?
- A. ibstat, then sminfo
- B. sminfo, then smpquery ND
- C. ibis, then ibsim
- D. sminfo, then smpquery NI
Answer : B
Question 3
An InfiniBand server stops working, and a system administrator runs the “ibstat” command that provides the following output:
What is the cause of the issue?
- A. The HCA port is faulty.
- B. There is no running SM in the fabric.
- C. The cable is disconnected.
- D. The neighboring switch port is faulty.
Answer : B
Question 4
After NCCL burn-in reports “transport retry count exceeded”, which corrective action addresses the underlying fabric issue?
- A. Switch from Ring to Tree algorithms via NCCL_ALGO=TREE
- B. Inspect InfiniBand link quality metrics (BER, symbol errors) and replace faulty cables
- C. Increase NCCL_IB_TIMEOUT to tolerate longer latencies
- D. Reduce message size to decrease network utilization
Answer : B
Question 5
An infrastructure engineer in an AI factory has successfully replaced a power supply unit on an NVIDIA DGX H100. After installation, both the IN and OUT LEDs on the new power supply illuminate solid green.
Which NVSM CLI command should the engineer use to quickly verify the overall system status and ensure it is operating as expected?
- A. nvsm show powermode
- B. nvsm show health
- C. nvsm show power
- D. nvsm show alerts
Answer : B
Question 6
After initial setup and health checks, the DGX H100 system administrator wants to verify that containers can access GPUs before running production workloads.
Which method is recommended for this validation?
- A. sudo docker run --gpus all --rm nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04 nvidia-smi
- B. sudo docker run --gpus all --rm nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04 systemctl
- C. sudo docker run --gpus all --rm nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04 ls-la
- D. sudo docker run –-rm nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04 nvidia-smi
Answer : A
Question 7
After ClusterKit reports “GPU-Host latency exceeds threshold”, which NVIDIA diagnostic tool should be used to isolate hardware faults?
- A. nvidia-smi topo –m to inspect GPU topology connections
- B. Re-run ClusterKit with –-stress=gpu –Y 60 to extend test duration
- C. DCGM Diags dcgmi diag –r 2
- D. ib_write_bw to measure InfiniBand bandwidth between nodes
Answer : C
Question 8
A media company is developing an AI platform for video content analysis that requires storing and processing large volumes of unstructured video data. The platform must support high throughput for data ingestion and provide efficient access for real-time analytics.
Given these requirements, which storage strategy should the company implement?
- A. Block storage for low latency and high performance
- B. Object storage for scalability and metadata management
- C. Tape storage for its cost-effectiveness and archival capabilities
- D. File storage for hierarchical organization and easy navigation
Answer : B
Question 9
A financial services firm is deploying an AI model for fraud detection that requires rapid inference and data retrieval across multiple sites.
Which feature should their storage system prioritize?
- A. Tape backup systems.
- B. Low-cost HDD solutions.
- C. High capacity with moderate speed.
- D. Multi-protocol data access with low latency.
Answer : D
Question 10
What command is needed to measure BER (Bit Error Rate)?
- A. mlxlink –d <device> -c –e
- B. mlxconfig –d <device> q
- C. msflint –d <device> q full
- D. ethtool –S<device>
Answer : A
Question 11
A System Administrator needs to change the scheduling behavior of a single GPU to use a fixed share scheduler.
What command achieves this?
- A. mixconfig –d /dev/mst/mt4123_pciconf0 set LINK_TYPE_P1=2
- B. esxcli –i 0 –mig 18
- C. nvidia-smi –i 0 –mig 1
- D. esxcli system module parameters set –m nvidia –p
Answer : D
Question 12
During cluster validation, the Cable Validation Tool (CVT) reports “Underperforming (BER)” for an InfiniBand link.
Which BER thresholds indicate a critical signal quality issue requiring cable replacement?
- A. Raw BER > 1e-6 or Effective BER> 1.5E-254 for ≤6hr measurements
- B. Effective BER > 0 during the first 125 minutes of link operation
- C. Temperature > 85 °C on transceiver module
- D. Rx power variance > 3dB between lanes
Answer : A
Question 13
During BCM cluster setup, an engineer must configure bonded network interfaces on DGX nodes for high availability.
Which cmsh command sequence properly configures a bond0 interface with two physical NICs?
-
A.
-
B.
-
C.
-
D.
Answer : C
Question 14
A company has a registered NGC account and their server has NCG CLI installed.
What step should be taken first to gain access to NGC?
- A. ngc config get
- B. ngc config update
- C. ngc config set
- D. ngc init
Answer : C
Question 15
You are installing the operating system as part of the initial setup for a new NVIDIA Base Command Manager (BCM) cluster.
Which two of the following actions are essential for a successful OS installation on the cluster’s head node? (Choose two.)
- A. Start the head node OS installation process with the system BIOS set to legacy boot mode instead of UEFI.
- B. Configure network switches for PXE boot to all compute nodes before installing the OS on the head node.
- C. Set the desired time zone and configure NTP synchronization during the OS installation wizard.
- D. Download the latest BCM ISO and verify its integrity using the provided checksum, then start the installation.
Answer : CD