NCP-AII試験無料問題集「NVIDIA AI Infrastructure 認定」

A. 'ngc system statue'

B. 'ngc registry model list'

C. 'ngc config validate'

D. 'ngc -version'

E. 'ngc setup verify'

A. The system memory is fully occupied. Deallocate some memory.

B. The application is using single precision floating-point operations. Switch to double precision to increase memory bandwidth utilization.

C. The GPU is being limited by power capping. Increase the power limit using 'nvidia-smi -pl' (if permitted) to allow the GPU to operate at higher clock speeds.

D. The application is using uncoalesced memory access patterns. Refactor the code to ensure contiguous memory access by threads within a warp.

E. The application is using a small transfer size per kernel launch. Increase the amount of data processed per kernel launch to amortize the overhead of kernel launch and data transfer.

A. Use a Dockerfile to create a container image with the NGC CLI pre-installed and configured.

B. Distribute the '-/.ngc/config.json' file to all hosts.

C. Use a configuration management tool like Ansible or Chef to automate the installation and configuration of the NGC CLI on all hosts.

D. Manually install the NGC CLI on each host, as automation is not recommended for security reasons.

E. Create a custom script that downloads the NGC CLI package, installs it using 'pip' , and configures the API key.

A. Using the annotation 'nvidia.com/gpu.product' and the value of the particular GPU device you desire.

B. Use a custom resource definition (CRD) to define MIG instances and reference it in the resource request.

C. Specify the MIG device profile name (e.g., 'nvidia.com/mig-l g.7gb: 1 in the resource request.

D. Use a generic resource request like 'nvidia.com/gpu: 0.125' to allocate a portion of the GPIJ.

E. Specify the full GPU ID (e.g., 'nvidia.com/gpu: 1') in the resource request.

A. iPerf/Netperf

B. Traceroute

C. Ping

D. tcpdump/Wireshark

E. netstat

A. 

B. 

C. 

D. 

E. 

A. The power supply is not providing enough power.

B. The system BIOS is limiting the power to the PCIe slot.

C. The driver is not correctly installed.

D. The GPU is faulty.

E. The GPIJ is in a low-power mode due to inactivity.

A. MIG hardware partitioning

B. vGPIJ scheduling

C. Kubernetes resource quotas

D. CUDA Multi-Process Service (MPS)

E. cgroups

A. Only testing the GPUs with a synthetic benchmark like 'linpack'.

B. Verifying that the server's power supply units (PSUs) are operating within their specified voltage and current ranges under full load.

C. Confirming the correct installation and functionality of all network interfaces (e.g., Ethernet, InfiniBand) relevant to the workload.

D. Ensuring the BIOS and BMC firmware are up to date.

E. Checking if the server's case fans are spinning.

A. The GPU is overheating, causing thermal throttling.

B. The version of the CUDA driver on the host is incompatible with the CUDA toolkit version used in the container.

C. The Docker container is not configured to use shared memory for data transfer with the GPU.

D. The application is performing frequent small memory transfers between the CPU and GPIJ.

E. The host machine's CPU is being heavily utilized, causing a bottleneck in data transfer to the GPU.

A. Using the 'ethtoor command to enable specific offload features like checksum offload, TCP segmentation offload (TSO), and UDP fragmentation offload (UFO).

B. Recompiling the Linux Kernel with the correct compilation flags.

C. Installing and configuring the appropriate Mellanox OFED drivers, which automatically enable many hardware offload features.

D. Running a custom script that programs the hardware offload engines directly.

E. Modifying the device tree to enable specific hardware features.

A. 100 Gigabit Ethernet

B. Fibre Channel

C. Gigabit Ethernet

D. 10 Gigabit Ethernet

E. InfiniBand

A. Check the DHCP client configuration to ensure it is enabled and properly configured to request IP addresses. Examine the logs for any errors.

B. Restart the networking service using 'systemctl restart networking'.

C. Manually assign static IP addresses to the interfaces using the 'ifconfig' command.

D. Re-flash the Bluefield OS image

E. Reinstall the Mellanox OFED drivers. A corrupted driver installation could cause network configuration issues.

A. MIG eliminates the need for GPU virtualization software.

B. MIG guarantees performance isolation between different workloads running on the same GPIJ.

C. MIG increases the overall FLOPS of the GPU.

D. MIG allows running multiple CUDA versions simultaneously on the same physical GPIJ.

E. MIG allows a single GPU to be partitioned into multiple isolated instances, improving resource utilization.

A. Incorrectly seated GPU in the PCle slot, leading to poor power delivery.

B. A failing voltage regulator module (VRM) on the GPU board, causing excessive power leakage.

C. Insufficient system RAM.

D. Degraded thermal paste between the GPU die and the heatsink.

E. A BIOS setting that is overvolting the GPU.

A. Implement InfiniBand as the interconnect technology for inter-server communication, utilizing appropriate InfiniBand cables and transceivers.

B. Replace existing Cat6 Ethernet cables with Cat8 cables.

C. Upgrade inter-server connections to the highest available Ethernet speed (e.g., from 100GbE to 400GbE) using appropriate transceivers and fiber optic cables.

D. Replace all existing transceivers with Active Optical Cables (AOCs).

E. Reduce the batch size of the LLM training job.

A. Examine the '/proc/driver/mlx4_core/versiorf file.

B. Use the 'mst status' command to query the device status and firmware version.

C. Run 'ethtool -i on a network interface associated with the DPIJ.

D. Query the DPIJ's BMC (Baseboard Management Controller) via IPMI or Redfish.

E. Check the system logs for firmware-related messages during boot.