A.  

Rx power variance > 3dB between lanes

B.  

Effective BER > 0 during the first 125 minutes of link operation

C.  

Raw BER > 1e-12 or Effective BER > 1.5E-254 for < 6hr measurements

D.  

Temperature > 85°C on transceiver module

A.  

High cable signal quality ensures that cable length and connector type do not play as big a role in deploying new infrastructure in the data center.

B.  

High cable signal quality minimizes bit error rates and supports reliable, high-throughput communication, reducing retransmissions and congestion across the network.

C.  

High cable signal quality reduces electromagnetic interference (EMI) and crosstalk, helping prevent unexpected packet drops during sustained workloads.

D.  

High cable signal quality enables effective use of Forward Error Correction (FEC), which is required for reliable operation at high data rates such as 200GbE and above.

A.  

Disable Secure Boot permanently in BIOS/UEFI settings.

B.  

Delete /etc/X11/xorg.conf to force driver reconfiguration.

C.  

Enroll the Machine Owner Key (MOK) during system reboot and enter the recorded password.

D.  

Reinstall drivers using apt-get install nvidia-driver-550 without rebooting.

A.  

Critical failure; expected is greater than 390 GB/s for HDR InfiniBand.

B.  

Suboptimal performance; requires FEC tuning to reach 380+ GB/s.

C.  

Optimal performance, indicating healthy fabric and GPUDirect RDMA.

D.  

Inconclusive; rerun with --stress=cpu to validate.

A.  

Optimal performance, indicating healthy fabric and GPUDirect RDM

A.  

B.  

Suboptimal performance; requires FEC tuning to reach 380+ GB/s.

C.  

Critical failure; expected is > 390 GB/s for HDR InfiniBand.

D.  

Inconclusive; rerun with --stress=cpu to validate.

A.  

Unlimited scalability by adding more DPUs without architectural changes.

B.  

Elimination of latency issues in data processing tasks.

C.  

Reduced CPU load by offloading data processing tasks to DPUs.

D.  

Enhanced I/O performance with NVMe storage access speeds.

A.  

A GPU tray upgrade failed.

B.  

A GPU is missing on the DGX system.

C.  

A GPU driver upgrade has failed.

D.  

The system has passed the hardware health check successfully.

A.  

Implement redundant switches with spanning tree protocol.

B.  

MLAG for bonded interfaces across redundant switches.

C.  

Use only one switch for all management and storage traffic.

D.  

Disable VLANs and use unmanaged switches.

A.  

QSA adapter.

B.  

SFP connectors.

C.  

SFP-to-1G BASE-T RJ45 adapter.

D.  

Standard QSFP-to-QSFP DAC cable.

A.  

Switch from Ring to Tree algorithms via NCCL_ALGO=TREE

B.  

Reduce message size to decrease network utilization

C.  

Increase NCCL_IB_TIMEOUT to tolerate longer latencies

D.  

Inspect InfiniBand link quality metrics (BER, symbol errors) and replace faulty cables

A.  

Sufficient networking, water-cooled racks, adequate rack power, sufficient storage, and rack space.

B.  

Sufficient storage, sufficient networking, adequate rack power, and compatible hardware.

C.  

Sufficient CPU capacity, PCIe slot allocation, sufficient cooling in the data center, and rack space.

D.  

Sufficient cooling in the data center, adequate rack power, compatible hardware, and PCIe slot allocation.

A.  

Navigate to ’Devices " > select a switch > " Cables ' tab to see ASIC firmware and transceiver versions.

B.  

Use " Topology’ view to visually inspect cable icons.

C.  

Run mlxlink -d lid- < LID > -m on each port manually.

D.  

Export all switch logs and grep for ’FW Version " .

A.  

Install latest OS images and drivers, confirm OS and container functionality, invite users for a monitored production trial, and collect workload feedback to plan any further diagnostics or updates.

B.  

Complete hardware and cabling, power on the system, update firmware and drivers, run full hardware health checks and stress diagnostics using NVSM, verify all GPU and system sensor logs, and validate GPU accessibility.

C.  

Update network topology, assign static IPs and DNS entries, register the system with NVIDIA, then conduct basic OS-level checks and enable user access after login testing is successful.

D.  

Power on the system, install all AI frameworks, configure the CUDA and library stack, set up user environments, then plan stress tests and diagnostics as part of ongoing routine operations.

A.  

A single VLAN for all types of network traffic.

B.  

Two networks: one for management and one for compute.

C.  

Four networks: compute, storage, out-of-band, and management.

A.  

Set up Active Directory and LDAP, configure role-based access controls and security settings first, install users, and skip network or hardware performance validation.

B.  

Perform application benchmarking first, use performance logs to identify bottlenecks, update switch and server firmware afterward, and then tune the network using performance tests.

C.  

Validate the software stack, test link connectivity and port health, run network benchmarks, run OSPF, ensure neighbors are exchanging route information, then stage AI workload tests.

D.  

Confirm switch and server firmware configuration, test link connectivity and port health, run network benchmarks, validate the software stack, then stage AI workload tests.

A.  

docker run --gpus ' " device=0,2 " ' nvidia/cuda:12.1-base nvidia-smi

B.  

docker run -e NVIDIA_VISIBLE_DEVICES=0,2 nvidia/cuda:12.1-base nvidia-smi

C.  

docker run --gpus all nvidia/cuda:12.1-base nvidia-smi -id=0,2

D.  

docker run --device /dev/nvidia0,/dev/nvidia2 nvidia/cuda:12.1-base nvidia-smi

A.  

Check /var/log/messages on the DPU operating system for update logs.

B.  

Query the DPU BMC with the Task ID of the installation process.

C.  

Power cycle the DPU immediately to force a rollback.

D.  

Run bfrec --status on the DPU to view flash progress.

A.  

export NGC_CLI_ORG=org-name & & ngc config set

B.  

ngc config list to manually edit the JSON configuration file.

C.  

ngc registry login --org org-name

D.  

ngc config set --org org-name --ace ace-name

A.  

The firmware is installed on the DPU.

B.  

The firmware is deleted from the DPU.

C.  

The firmware is copied to the DPU but not installed.

D.  

The firmware is waiting on reboot to become active.

A.  

Perform a software-driven restart on the operating system of every compute node, then use advanced tools to check firmware status and reissue update commands if any firmware appears inactive afterward.

B.  

Initiate the required cold reset or power cycle to activate updated firmware, reset the BMC using the recommended command, and perform an AC power cycle when required for EROT and CPLD firmware activation.

C.  

Initiate a cold power cycle on all node trays to activate firmware, follow with a DGX reboot procedure, and use the management interface to finish activating CPLD firmware on the host.

D.  

Execute a single operating system reboot on the DGX after the update process, then reset the software stack and verify status using diagnostic commands on each node.

A.  

Use NCCL_TESTS_SPLIT= " DIV 8 " with point-to-point tests

B.  

Use all_reduce_perf -b 8 -e 16G -f 2 -g 8 with NCCL_TESTS_SPLIT= " AND 0x1 "

C.  

Use reduce_scatter_perf -b 8 -e 16G -f 2 -g 4

D.  

Use all_reduce_perf -b 8 -e 16G -f 2 -g 8 without splits

A.  

ngc config get

B.  

ngc init

C.  

ngc config set

D.  

ngc config update

A.  

To guarantee that hardware inventory tools can report serial numbers and manufacturer codes for asset management, which is critical for future support and troubleshooting.

B.  

To ensure stability, bandwidth, and compatibility across the cluster, avoiding link issues and performance loss.

C.  

To allow the network operating system to automatically discover all connected transceivers with heterogeneous firmware.

D.  

To reduce GPU memory consumption during distributed training jobs.

A.  

mig-parted

B.  

nvidia-smi

C.  

nvcc

D.  

nvlink-config

A.  

Use a single VLAN for both Out-of-Band management and compute fabric to simplify network design.

B.  

Configure Out-of-Band management interfaces to be accessible from any subnet within the data center for maximum flexibility.

C.  

Connect Out-of-Band management ports to the same switch as user traffic for easier troubleshooting.

D.  

Place all BMC and management interfaces on an isolated Out-of-Band network with access restricted by firewall rules.

A.  

Run a deep learning workload to stress test the GPUs and check whether the issue persists.

B.  

Check the NVIDIA System Management Interface (nvidia-smi) for GPU status and temperatures.

C.  

Power drain then restart the DGX and check if the performance degradation resolves.

D.  

Increase the fan speed to maximum and check whether the performance improves.

A.  

Repeated CRC errors and intermittent port flapping reported by switch counters.

B.  

Output of ifconfig showing link speed at the expected rate on both ends of the cable.

C.  

Network pings between all cluster nodes return responses with delays under 2 ms on a 100Gb network.

A.  

Confirm the system fans are running at 100% under all workloads to prevent overheating.

B.  

Review the system BIOS to ensure GPU overclocking is enabled for maximum performance.

C.  

Use NVSM to disable unused PCIe devices to reduce overall system heat output.

D.  

Verify that each DGX system is connected to redundant, properly rated PDUs and that all power supplies are reporting nominal input.

A.  

lspci | grep NVIDIA

B.  

nvidia-smi

C.  

nvidia-gpu-status

D.  

iblinkinfo

A.  

Inconclusive; rerun with point-to-point tests.

B.  

Optimal performance; bus bandwidth near theoretical peak for NDR InfiniBand.

C.  

Critical failure; bus bandwidth exceeds hardware capabilities.

D.  

Suboptimal performance; algorithm bandwidth should match bus bandwidth.

A.  

Use the command sudo mlxconfig -d /dev/mst/ < device > set LINK_TYPE_P1=2 to enable Ethernet on the Bluefield-3 devices.

B.  

Use the command sudo mlxconfig -d /dev/mst/ < device > set DISABLE_SPECTRUM_X=1 to reduce overhead.

C.  

Use the command sudo mlxconfig -d /dev/mst/ < device > set INTERNAL_CPU_OFFLOAD_ENGINE=1 to configure the SuperNIC to operate in NIC mode.

D.  

Use the command sudo mlxconfig -d /dev/mst/ < device > set DPU_MODE=1 to set up the Bluefield-3 devices in DPU mode.

A.  

nvidia-smi -q | grep " GPU Stress Test "

B.  

sudo nvsm stress-test --force

C.  

stress --cpu $(nproc) --io $(nproc) --timeout 600

D.  

./gpu_burn 60

A.  

The HCA port is faulty.

B.  

There is no running SM in the fabric.

C.  

The neighboring switch port is faulty.

D.  

The cable is disconnected.

A.  

Simple ring topology connecting all nodes in a loop.

B.  

Fat-Tree topology with rail-optimized design.

C.  

Single flat Ethernet network for all traffic.

D.  

Star topology with all nodes connected to a single central switch.

A.  

The network card has no link / connection.

B.  

A boot disk has failed.

C.  

Multiple GPUs have failed.

D.  

There are more than two failed power supplies.

A.  

Assign a floating Virtual IP address that can automatically migrate between the primary and secondary head nodes during failover.

B.  

Compute nodes must be powered on and performing work to initiate synchronization of the head nodes.

C.  

After configuration is complete, simulate a failover by stopping BCM services on the active head node to verify that all services are running on the secondary node with no interruption.

D.  

Configure both head nodes to use independent static IP addresses for BCM services instead of relying on a shared virtual IP address.

E.  

During configuration, explicitly synchronize both the configuration and state data directories from the primary to the secondary head node to ensure consistency.