최신 NCP-AII 무료덤프 - NVIDIA AI Infrastructure

문제1

After replacing a GPU in a multi-GPU server, you notice that the new GPU is consistently running at a lower clock speed than the other GPUs, even under load. *nvidia-smi' shows the 'Pwr' state as 'P8' for the new GPU, while the others are at 'PO'. What is the MOST probable cause?

A. The new GPU is not receiving sufficient power; check the power connections and PSU capacity.

B. The new GPU is overheating and throttling performance.

C. The driver is not properly recognizing the new GPU's capabilities; reinstall the driver.

D. The new GPU is a lower-performance model than the other GPUs.

E. The new GPU requires a firmware update that hasn't been applied.

정답: A

설명: (DumpTOP 회원만 볼 수 있음)

문제2

Consider a scenario where you need to run two different deep learning models, Model A and Model B, within separate Docker containers on the same NVIDIA GPU. Model A requires CUDA 11.2, while Model B requires CUDA 11.6. How can you achieve this while minimizing conflicts and ensuring each model has access to its required CUDA version?

A. Mount the CUDA libraries from the host machine into both containers using Docker volumes, ensuring each container has access to both CUDA versions.

B. Use separate Docker images for each model, each based on the appropriate 'nvidia/cuda' image (e.g., 'nvidia/cuda:ll .2-base-ubuntu20.04' and nvidia/cuda: 1 1.6-base-ubuntu20.04 s).

C. Install both CUDA 11.2 and CUDA 11.6 inside each Docker container and use 'LD LIBRARY PATH' to switch between the CUDA versions for each model.

D. Create a single Docker image with both CUDA versions and dynamically link the correct CUDA libraries at runtime using environment variables.

E. Install both CUDA 11.2 and CUDA 11.6 on the host system and use 'CUDA VISIBLE DEVICES* to isolate each model to a specific CUDA version.

정답: B

설명: (DumpTOP 회원만 볼 수 있음)

문제3

You observe high latency and low bandwidth between two GPUs connected via an NVLink switch. You suspect a problem with the NVLink link itself. Which of the following methods would be the most effective in diagnosing the physical NVLink link health?

A. Using 'ping' to check network connectivity between the servers.

B. Using 'iperf3' to measure network throughput between the servers.

C. Running a CUDA-aware memory bandwidth test specifically designed for NVLink.

D. Examining system logs for NVLink-related error messages.

E. Physically inspecting the NVLink cables for damage.

정답: C,D,E

설명: (DumpTOP 회원만 볼 수 있음)

문제4

You want to automate the NGC CLI installation process across multiple hosts in your infrastructure. What are the best practices to achieve this?

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

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

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

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

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

정답: B,D,E

설명: (DumpTOP 회원만 볼 수 있음)

문제5

Consider the following *iptables' rule used in an A1 inference server. What is its primary function?
iptables -A INPUT -p tcp --dport 8080 -j ACCEPT

A. Drops all IJDP traffic destined for port 8080.

B. Blocks all TCP traffic on port 8080.

C. Accepts all TCP traffic destined for port 8080.

D. Accepts all TCP traffic originating from port 8080.

E. Redirects TCP traffic from port 8080 to another port.

정답: C

설명: (DumpTOP 회원만 볼 수 있음)

문제6

When installing multiple NVIDIA GPUs, which of the following factors are MOST important to consider regarding PCIe slot configuration?
(Choose two)

A. Ensure all GPUs are installed in slots of the same color.

B. Ensure each GPU is installed in a slot with sufficient PCIe lanes (e.g., x16).

C. Ensure the PCIe slots are directly connected to the CPU for optimal bandwidth.

D. Ensure all GPUs have the same PCIe generation (e.g. Gen4).

E. Install the GPUs in the lowest numbered slots first.

정답: B,C

설명: (DumpTOP 회원만 볼 수 있음)

문제7

A large language model (LLM) training job is running across multiple NVIDIAAI 00 GPUs in a cluster. You observe that the GPUs within a single server are communicating efficiently via NVLink, but the inter-server communication over Ethernet is becoming a bottleneck. Which of the following strategies, focusing on cable and transceiver selection, would MOST effectively address this inter-server communication bottleneck? (Choose TWO)

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

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

C. Replace existing Cat6 Ethernet cables with Cat8 cables.

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

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

정답: A,B

설명: (DumpTOP 회원만 볼 수 있음)

문제8

You are deploying an NVIDIA-Certified A1 server. The documentation specifies a minimum airflow requirement for the GPUs. How would you BEST monitor the GPU temperatures and ensure the airflow is adequate during a stress test?

A. Measure the ambient temperature around the server.

B. Use a software utility like 'psensor' to monitor GPU temperature.

C. Use IPMI sensors to monitor GPU temperature and fan speeds.

D. Use a thermal camera to measure the GPU heatsink temperature.

E. Use 'nvidia-smi' to monitor GPU temperature and visually inspect the fans.

정답: C

설명: (DumpTOP 회원만 볼 수 있음)

문제9

You're optimizing an AMD EPYC server with 4 NVIDIAAIOO GPUs for a large language model training workload. You observe that the GPUs are consistently underutilized (50-60% utilization) while the CPUs are nearly maxed out. Which of the following is the MOST likely bottleneck?

A. The storage system (SSD/NVMe) is too slow, leading to data starvation.

B. The NCCL (NVIDIA Collective Communications Library) is not properly configured for inter-GPU communication.

C. Insufficient CPU cores to prepare and feed data to the GPUs.

D. The system RAM is too small, causing excessive swapping.

E. The PCle interconnect between the CPUs and GPUs is saturated.

정답: C

설명: (DumpTOP 회원만 볼 수 있음)

문제10

You are attempting to install NGC CLI on a CentOS 7 system, but the 'pip install nvidia-cli' command fails with a 'Could not find a version that satisfies the requirement nvidia-cli' error. You have confirmed that 'pip' is installed and working. What could be the cause of this issue?

A. The system's firewall is blocking access to the Python Package Index (PyPl). CentOS 7 is not supported by NGC CLI.

B. The system's package manager (YUM) is not configured correctly, preventing 'pip' from finding the NGC CLI package.

C. The 'pip' version is outdated and incompatible with the NGC CLI package. Upgrade 'pip' using 'pip install -upgrade pip'.

D. The CentOS 7 system does not have the required Python version installed. NGC CLI requires Python 3.6 or later.

정답: C,D

설명: (DumpTOP 회원만 볼 수 있음)

문제11

You are observing that the memory bandwidth being achieved by your CUDA application on an NVIDIAAIOO GPU is significantly lower than the theoretical peak bandwidth. Which of the following could be potential causes for this, and what actions can you take to validate or mitigate them? (Select all that apply)

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

B. 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.

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

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

E. 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.

정답: A,B,E

설명: (DumpTOP 회원만 볼 수 있음)

문제12

You're optimizing an Intel Xeon server with 4 NVIDIAAIOO GPUs for a computer vision application that uses CODA. You notice that the GPU utilization is fluctuating significantly, and performance is inconsistent. Using 'nvprof, you identify that there are frequent stalls in the CUDA kernels due to thread divergence. What are possible causes and solutions?

A. The CUDA compiler is generating suboptimal code. Try using different compiler optimization flags (e.g., '-O3' or '-ftz=true').

B. The CUDA code contains conditional branches that lead to different execution paths for different threads within the same warp. Rewrite the CUDA code to minimize branching and favor uniform execution paths within warps.

C. The GPUs are overheating, causing thermal throttling. Improve the server's cooling.

D. The CUDA driver version is incompatible with the CUDA toolkit version. Update the CUDA driver to a compatible version.

E. The input data is not properly aligned in memory. Ensure that data is aligned to 128-byte boundaries using aligned memory allocation techniques.

정답: B,E

설명: (DumpTOP 회원만 볼 수 있음)

문제13

You are designing an Ai server infrastructure using NVIDIA HGX AIOO modules. The server's power supply units (PSUs) are configured in a redundant (N+1) setup. The individual PSUs are rated for 3000W each, and the server contains three PSUs. If the expected peak power consumption of the HGX A100 modules and other components is 5500W, what is the safety margin (in Watts) in the power budget?

A. 500W

B. 3500W

C. 0W

D. 9000W

E. 5000W

정답: A

설명: (DumpTOP 회원만 볼 수 있음)

최신 NCP-AII 무료덤프 - NVIDIA AI Infrastructure

우리와 연락하기

유용한 링크

최신 업데이트