[2026] Pass NCP-AII Exam - Real Questions & Answers [Q28-Q52]

[2026] Pass NCP-AII Exam - Real Questions and Answers

NCP-AII Exam Questions Get Updated [2026] with Correct Answers

NEW QUESTION # 28
You are deploying a multi-node NVIDIA GPU cluster for distributed deep learning. Each node has a different ambient operating temperature due to varying airflow patterns within the data center. To ensure optimal performance and longevity of the GPUs across all nodes, which approach is MOST effective for managing GPU power limits?

• A. Manually adjust the fan speeds of each GPU to ensure they are all running at maximum RPM.
• B. Rely on the default power management settings provided by the GPU driver.
• C. Implement dynamic power management using NVIDIA's Data Center GPU Manager (DCGM) to adjust power limits on a per-GPU basis, taking into account real- time temperature readings and workload characteristics.
• D. Disable power capping altogether to allow GPUs to operate at their maximum potential performance.
• E. Set a uniform power limit for all GPIJs across the entire cluster based on the GPU's Thermal Design Power (TDP) specification.

Answer: C

Explanation:
Option C, using DCGM for dynamic power management, is the most effective approach. It allows for per-GPU power limit adjustments based on real-time conditions, optimizing performance while ensuring thermal safety and longevity across nodes with different operating temperatures. A uniform power limit (A) might be too restrictive for some nodes or insufficient for others. Disabling power capping (B) risks overheating and damage. Default settings (D) may not be optimal. Manually adjusting fan speeds (E) can help, but doesn't address power limits directly.

NEW QUESTION # 29
You're working with a large dataset of microscopy images stored as individual TIFF files. The images are accessed randomly during a training job. The current storage solution is a single HDD. You're tasked with improving data loading performance. Which of the following storage optimizations would provide the GREATEST performance improvement in this specific scenario?

• A. Replacing the HDD with a RAID 5 array of HDDs.
• B. Compressing the TIFF files using a lossless compression algorithm.
• C. Implementing data deduplication on the storage volume.
• D. Replacing the HDD with a single NVMe SS
• E. Migrating the data to a large, sequential HDD.

Answer: D

Explanation:
Random access to numerous small files is a classic use case where SSDs excel due to their low latency. Replacing the HDD with an NVMe SSD (option D) will provide the most significant performance improvement. Data deduplication (A) saves storage space but doesn't directly improve random access speed. Migrating to a sequential HDD (B) is counterproductive for random access. RAID 5 (C) provides some performance improvement but not as much as an SSD. Compression (E) can reduce storage space but adds overhead during decompression.

NEW QUESTION # 30
Which of the following statements are true regarding the use of Congestion Management (CM) and Congestion Avoidance (CA) techniques within an InfiniBand fabric using NVIDIA technology? (Select TWO)

• A. InfiniBand's Explicit Congestion Notification (ECN) is a CA mechanism that allows switches to signal congestion to endpoints before packet loss occurs.
• B. CM/CA mechanisms are primarily implemented at the IP layer and are independent of the InfiniBand transport layer.
• C. CM/CA are not relevant in InfiniBand fabrics because InfiniBand's lossless nature guarantees that no packets will ever be dropped due to congestion.
• D. CM can include techniques like rate limiting to throttle traffic flows when congestion is detected.
• E. CM aims to reduce the severity of congestion once it has already occurred, while CA aims to prevent congestion from happening in the first place.

Answer: A,D,E

Explanation:
CM and CA are crucial for maintaining performance in InfiniBand fabrics. CM addresses congestion that has already occurred, while CA tries to prevent it. ECN is a key CA mechanism in InfiniBand. InfiniBand is lossless but experiences performance degradation if congestion occurs; CM helps mitigate this. They are not implemented at the IP layer; they are integral parts of the InfiniBand transport. Rate limiting is a viable CM strategy.

NEW QUESTION # 31
A security policy requires you to log all NGC CLI commands executed on a specific host. How can you achieve this without modifying the NGC CLI source code?

• A. NGC CLI automatically logs all commands to a file in the '/var/log/ngc' directory.
• B. Create a wrapper script around the 'ngc' command that logs the command and its arguments before executing the actual NGC CLI command.
• C. Implement system-level auditing using tools like 'auditd' (Linux) or Windows Event Logging to capture all executed commands, including NGC CLI commands.
• D. Configure the NGC CLI to send all commands to a remote syslog server.
• E. Use the 'history' command in Linux to view the command history for the user account used to execute the NGC CLI commands.

Answer: B,C

Explanation:
System-level auditing (B) provides a comprehensive way to log all executed commands. Creating a wrapper script (C) allows you to log the command before it's executed. The 'history' command (A) only captures commands executed in the current shell session. Option D is incorrect as NGC CLI does not log automatically to this directory, and option E is not a supported feature.

NEW QUESTION # 32
Consider the following simplified CUDA code snippet intended to perform a vector addition:
What are critical steps to validate that this code is correctly utilizing the GPU hardware and producing accurate results?

• A. All of the above
• B. After the 'cudaMemcpy' from device to host, compare the 'c' array to a CPU-based vector addition result to verify correctness.
• C. Use 'cudaDeviceSynchronize()' after launching the kernel to ensure the GPU computations are complete before copying data back to the host.
• D. Run the code with 'cuda-memchecK to detect memory access errors or race conditions.
• E. Ensure that 'blocksPerGrid' and 'threadsPerBlock' are appropriately chosen to maximize GPU occupancy, potentially using the NVIDIA CUDA Occupancy Calculator.

Answer: A

Explanation:
All options highlight important validation steps. Optimizing block/thread configuration maximizes GPU utilization. Verifying results against a CPU-based calculation ensures correctness. 'cudaDeviceSynchronize()' guarantees GPU computations are finished before data transfer. 'cuda-memcheck' detects memory errors. Failing to do any of these could lead to subtle errors or performance bottlenecks.

NEW QUESTION # 33
You are using the NVIDIA Container Toolkit in a Kubernetes environment with multiple GPUs per node. You want to ensure that pods can request specific GPUs on a node, rather than simply requesting 'any' GPU. Which Kubernetes feature, in conjunction with the NVIDIA Device Plugin, allows you to achieve this fine-grained GPU resource allocation?

• A. Node Affinity
• B. Topology Manager
• C. Resource Quotas
• D. Device Plugins API
• E. Taints and Tolerations

Answer: B

Explanation:
The Kubernetes Topology Manager (C) allows you to align resource allocations (including GPIJs) with specific NIJMA nodes. This is critical for performance when dealing with multiple GPUs per node. Resource Quotas (A) limit resource usage but don't control specific GPU selection. Node Affinity (B) selects nodes based on labels, not specific GPUs. The Device Plugins API (D) enables GPIJ discovery, but the Topology Manager is needed for fine-grained allocation within a node. Taints and Tolerations (E) are used to prevent pods from being scheduled on certain nodes unless they have the corresponding toleration, and does not directly allow for the selection of a particular GPU.

NEW QUESTION # 34
You're optimizing an Intel Xeon server with 4 NVIDIA GPUs for inference serving using Triton Inference Server. You've deployed multiple models concurrently. You observe that the overall throughput is lower than expected, and the GPU utilization is not consistently high.
What are potential bottlenecks and optimization strategies? (Select all that apply)

• A. Model loading and unloading overhead. Use model ensemble or dynamic batching to reduce frequency.
• B. The models are memory-bound. Reduce the model precision (e.g., FP32 to FP16 or INT8).
• C. The GPUs are underutilized due to small batch sizes. Implement dynamic batching to increase batch sizes.
• D. Insufficient CPU cores to handle the model loading and preprocessing requests. Increase the number of Triton instance groups for CPU-based models.
• E. Insufficient PCle bandwidth between CPU and GPIJs. Reconfigure PCle lanes to improve bandwidth allocation to each GPIJ.

Answer: A,B,C,D

Explanation:
Multiple factors can contribute to low throughput in inference serving. Model loading overhead is significant, and dynamic batching is crucial to maximize throughput. Insufficient CPU cores and memory constraints on the GPU also limit performance. Model precision reduction helps reduce memory footprint and increase throughput. While PCle bandwidth is a factor, it is often not the primary bottleneck in inference serving.

NEW QUESTION # 35
You want to automate the NGC CLI installation process across multiple hosts in your infrastructure. What are the best practices to achieve this?

• A. Distribute the '-/.ngc/config.json' file to all hosts.
• B. Create a custom script that downloads the NGC CLI package, installs it using 'pip' , and configures the API key.
• 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. Use a Dockerfile to create a container image with the NGC CLI pre-installed and configured.

Answer: B,C,E

Explanation:
Automation is highly recommended. Configuration management tools (A), custom scripts (B), and containerization (D) are all viable options for automating the NGC CLI installation process. Manually installing on each host is inefficient and error-prone. Distributing the config.json (E) could be a security risk.

NEW QUESTION # 36
What is the primary function of the NVIDIA Container Toolkit, and how does it facilitate the use of GPUs within containerized environments? (Multiple Answers)

• A. It manages the lifecycle of containers running GPU-accelerated workloads.
• B. It enables monitoring of GPU utilization within containers.
• C. It allows containers to access and utilize NVIDIA GPUs by injecting the necessary drivers and libraries into the container runtime environment.
• D. It automatically installs the necessary NVIDIA drivers inside the container.
• E. It provides a set of command-line tools for managing NVIDIA drivers on the host system.

Answer: B,C

Explanation:
The NVIDIA Container Toolkit allows containers to access and utilize NVIDIA GPUs by injecting the necessary drivers and libraries into the container runtime environment and It enables monitoring of GPU utilization within containers. While it requires proper drivers to be installed, the toolkit does not manage host drivers directly. The NVIDIA container toolkit relies on container runtimes, and container runtimes manage the container lifecycle. The container toolkit does not automatically install drivers inside containers.

NEW QUESTION # 37
You need to verify the NVLink connectivity between GPUs in a DGX server. Which command-line utility is the MOST reliable and provides detailed NVLink status?

• A. Ispci
• B. gpustat
• C. dcgmi diag -t 1004
• D. nvidia-smi
• E. nvlink_info (Hypothetical command)

Answer: C

Explanation:
'dcgmi diag -t 1004' is the correct command. 'nvidia-smi' provides basic GPIJ information, but 'dcgmi diag -t 1004' (part of the Data Center GPU Manager) provides specific diagnostic tests for NVLink connectivity. 'Ispci' lists PCle devices, not specifically NVLink. 'gpustat' is a monitoring tool. 'nvlink_info' is hypothetical.

NEW QUESTION # 38
You are tasked with ensuring optimal power efficiency for a GPU server running machine learning workloads. You want to dynamically adjust the GPU's power consumption based on its utilization. Which of the following methods is the MOST suitable for achieving this, assuming the server's BIOS and the NVIDIA drivers support it?

• A. Disable ECC (Error Correcting Code) on the GPU to reduce power consumption.
• B. Configure the server's BIOS/UEFI to use a power-saving profile, which will automatically reduce the GPU's power consumption when idle.
• C. Enable Dynamic Boost in the NVIDIA Control Panel (if available), which will automatically allocate power between the CPU and GPU based on their current needs.
• D. Use NVIDIA's Data Center GPU Manager (DCGM) to monitor GPU utilization and dynamically adjust the power limit based on a predefined policy.
• E. Manually set the GPU's power limit using 'nvidia-smi -pl and create a script to monitor utilization and adjust the power limit periodically.

Answer: D

Explanation:
DCGM provides the most comprehensive and automated solution for dynamic power management. It can monitor GPIJ utilization in real-time and adjust the power limit based on predefined policies, ensuring optimal power efficiency without manual intervention. Manually adjusting the power limit is possible but requires scripting and continuous monitoring. Dynamic Boost is typically for laptops, and BIOS power profiles may not be fine-grained enough. Disabling ECC reduces power but compromises data integrity.

NEW QUESTION # 39
Consider the following scenario: You have a BlueField-2 DPU installed in a server. You are trying to establish RDMA (Remote Direct Memory Access) communication between the DPU and another server. However, the RDMA connection fails. Which of the following is the most crucial factor to verify in this scenario?

• A. That the power supply to the DPU is providing sufficient wattage.
• B. That the clocks on both servers are synchronized using NTP (Network Time Protocol).
• C. That the TCP window size is properly tuned for high-bandwidth communication.
• D. That the correct MTU (Maximum Transmission Unit) is configured on the network interfaces involved in the RDMA connection.
• E. That the appropriate RDMA kernel modules (e.g., , 'ib_uverbs) are loaded on both the DPUand the remote server.

Answer: E

Explanation:
For RDMA to function correctly, the necessary kernel modules must be loaded on both ends of the connection. These modules provide the core functionality for RDMA communication. While MTU and other network settings are important, the absence of the RDMA modules will prevent the connection from being established in the first place. Clock synchronization and TCP window size are less directly related to the initial RDMA connection failure. Power supply issues would likely manifest in other ways.

NEW QUESTION # 40
You are troubleshooting a performance issue with a GPU-accelerated application running inside a Docker container. The 'nvidia-smi' output inside the container shows the GPU is being utilized, but the performance is significantly lower than expected. Which of the following could be the cause of this performance bottleneck?

• A. The host machine's CPU is being heavily utilized, causing a bottleneck in data transfer to the GPU.
• B. The application is performing frequent small memory transfers between the CPU and GPIJ.
• C. The version of the CUDA driver on the host is incompatible with the CUDA toolkit version used in the container.
• D. The Docker container is not configured to use shared memory for data transfer with the GPU.
• E. The GPU is overheating, causing thermal throttling.

Answer: A,B,C,E

Explanation:
Several factors could contribute to reduced GPU performance within a Docker container, even if the GPU is being utilized. A heavily loaded CPU (A) can bottleneck data transfer to the GPU. Incompatible CUDA driver versions between host and container (C) cause unexpected errors, and CUDA drivers are important for GPU support. Frequent small memory transfers between CPU and GPU (D) can be inefficient. Overheating (E) can cause the GPU to throttle its performance. While shared memory optimization (B) can help, it's not always the primary cause of the initial performance drop.

NEW QUESTION # 41
Your AI training pipeline involves a pre-processing step that reads data from a large HDF5 file. You notice significant delays during this step. You suspect the HDF5 file structure might be contributing to the slow read times. What optimization technique is MOST likely to improve read performance from this HDF5 file?

• A. Storing the HDF5 file on a network file system like NFS.
• B. Compressing the HDF5 file using gzip.
• C. Reorganizing the HDF5 file to improve data contiguity and chunking.
• D. Converting the HDF5 file to a CSV file.
• E. Encrypting the HDF5 file for enhanced security.

Answer: C

Explanation:
Reorganizing the HDF5 file (option C) to improve data contiguity and chunking is the most effective optimization. HDF5 performance is highly dependent on how the data is laid out within the file. Contiguous data and optimal chunk sizes allow for more efficient 1/0 operations. Converting to CSV (A) loses the hierarchical structure of HDF5. Storing on NFS (B) adds network overhead. Compression (D) can reduce storage space but increases decompression overhead. Encryption (E) adds overhead without improving read performance.

NEW QUESTION # 42
You have a server with two NVIDIA GPUs connected via NVLink. You want to verify that NVLink is functioning correctly. Which command(s) or tool(s) can you use to check the NVLink status and bandwidth?

• A. 'nvidia-smi nvlink -statue
• B. 'nvidia-settings' (GUI tool)
• C. 'nvidia-smi topo -m'
• D. Ispci'
• E. nvcc -version'

Answer: A,C

Explanation:
'nvidia-smi nvlink -statuS provides a direct overview of the NVLink status, including link speed and errors. 'nvidia-smi topo shows the topology of the GPUs and how they are connected, including NVLink connections. 'Ispci' lists PCl devices but doesn't provide NVLink- specific information. 'nvcc -version' checks the CUDA compiler version. 'nvidia-settings' is a GUI tool that can display some information, but it's less precise than 'nvidia-smr for NVLink status.

NEW QUESTION # 43
During HPL execution on a DGX cluster, the benchmark fails with "not enough memory" errors despite sufficient physical RAM. Which HPL.dat parameter adjustment is most effective?

• A. Reduce the problem size while maintaining the same block size.
• B. Disable double-buffering via BCAST parameter.
• C. Increase block size to 6144 to maximize GPU utilization.
• D. Set PMAP to 1 to enable process mapping.

Answer: A

Explanation:
High-Performance Linpack (HPL) is a memory-intensive benchmark that allocates a large portion of available GPU memory to store the matrix $N$. While a server may have 2TB of physical system RAM, the "not enough memory" error usually refers to theHBM (High Bandwidth Memory)on the GPUs themselves. In a DGX H100 system, each GPU has 80GB of HBM3. If the problem size ($N$) specified in the HPL.dat file is too large, the required memory for the matrix will exceed the aggregate capacity of the GPU memory.
Reducing the problem size ($N$) while maintaining the optimal block size ($NB$) ensures that the problem fits within the GPU memory limits while still pushing the computational units to their peak performance.
Increasing the block size (Option C) would actually increase the memory footprint of certain internal buffers, potentially worsening the issue. Reducing $N$ is the standard procedure to stabilize the run during the initial tuning phase of an AI cluster bring-up.

NEW QUESTION # 44
An AI server with 8 GPUs is experiencing random system crashes under heavy load. The system logs indicate potential memory errors, but standard memory tests (memtest86+) pass without any failures. The GPUs are passively cooled. What are the THREE most likely root causes of these crashes?

• A. Insufficient airflow within the server, leading to overheating of the GPUs and VRMs.
• B. A faulty power supply unit (PSU) that is unable to provide stable power under peak load.
• C. Incompatible NVIDIA driver version with the installed Linux kernel.
• D. GPIJ memory errors that are not detectable by standard CPU-based memory tests.
• E. Network congestion causing intermittent data corruption during distributed training.

Answer: A,B,D

Explanation:
GPU memory errors (B) are a strong possibility, as CPU-based tests don't test GPU memory directly. Insufficient airflow (C) is likely due to the passive cooling, leading to thermal instability. A faulty PSU (D) can cause random crashes under load due to power fluctuations. Driver incompatibility (A) is less likely to cause random crashes after initial setup, and network congestion (E) usually results in training slowdowns rather than system crashes.

NEW QUESTION # 45
You are following the official steps to install the NVIDIA Container Toolkit using a package manager on Ubuntu. After importing the NVIDIA package repository and GPG key, what is the next action?

• A. Format the disk to clear any existing NVIDIA-related dependencies first.
• B. Reboot the host system to apply the repository changes and proceed.
• C. Download the CUDA toolkit installer from NVIDIA'S official website.
• D. Install the nvidia-container-toolkit package using your package manager.

Answer: D

Explanation:
The NVIDIA Container Toolkit (formerly nvidia-docker2) is the essential middleware that allows Docker, Podman, or Containerd to "see" and utilize the host's GPU hardware. The standard installation workflow on Debian-based systems like Ubuntu involves three core phases: repository configuration, package installation, and runtime configuration. Once the GPG key is added (to ensure package integrity) and the .list file is placed in /etc/apt/sources.list.d/ (to point to the NVIDIA production servers), the local package index must be refreshed via apt-get update. Immediately following this, the administrator must install the toolkit using the command sudo apt-get install -y nvidia-container-toolkit. Rebooting (Option A) is unnecessary at this stage because no kernel modules have been modified yet. Downloading the CUDA Toolkit (Option D) is a separate step; notably, the Container Toolkit allows containers to run CUDA applications even if the host only has the NVIDIA driver installed, making the driver-not the host CUDA toolkit-the primary prerequisite.

NEW QUESTION # 46
Consider the following Python code snippet which attempts to extract Digital Optical Monitoring (DOM) data from a transceiver using a hypothetical library 'transceiver_utils'. The transceiver is connected to port 'eth0'. However, the code consistently throws a 'TransceiverError: Invalid port' exception. What is the MOST likely cause of this error?

• A. The fiber cable connected to the transceiver is damaged.
• B. The port 'eth0' does not exist or is not correctly associated with the transceiver.
• C. The 'transceiver_utils' library is outdated and does not support DOM data extraction.
• D. The transceiver does not support DOM functionality.
• E. The Python code requires root privileges to access transceiver data.

Answer: B

Explanation:
The 'Invalid port' error strongly suggests that the specified port identifier ('eth0') is either incorrect or not properly linked to the transceiver by the operating system or networking stack. While other issues like outdated libraries, lack of DOM support, or cable damage could cause problems, the specific error message points directly to a port configuration issue.

NEW QUESTION # 47
Consider a scenario where you're using GPUDirect Storage to enable direct memory access between GPUs and NVMe drives. You observe that while GPUDirect Storage is enabled, you're not seeing the expected performance gains. What are potential reasons and configurations you should check to ensure optimal GPUDirect Storage performance? Select all that apply.

• A. Check if the file system supports direct I/O (e.g., using 'directio' mount option).
• B. Ensure that the NVMe drives are connected to the system via PCle Gen4 or Gen5.
• C. Confirm that the CUDA driver version is compatible with GPIJDirect Storage.
• D. Verify that the NVMe drives are properly configured in a RAID 0 configuration.
• E. Disable CPU-side caching to force all I/O operations to go directly to the GPU memory.

Answer: A,B,C

Explanation:
Explanation:GPUDirect Storage requires PCle Gen4/Gen5 for sufficient bandwidth (B). The CUDA driver must be compatible with GPUDirect Storage (C). Direct I/O support in the file system is essential to bypass the OS cache and allow direct GPU access (D). RAID 0 (A) is about storage speed but not directly related to GDS functionality. Disabling CPU-side caching (E) is usually detrimental as it can reduce overall system performance. Note, this is not always bad but needs to be tested depending on application.

NEW QUESTION # 48
You're setting up a cluster with 8 NVIDIA A100 GPUs. Each GPU needs to read 4GB/s from storage to keep it fully utilized. The network connecting the storage and compute nodes has a bandwidth of 25GB/s. What is the maximum number of GPUs that can be simultaneously saturated with data without exceeding the network bandwidth?

• A. 0
• B. 1
• C. 2
• D. 3
• E. 4

Answer: E

Explanation:
Each GPU requires 4GB/s of bandwidth. The network has a bandwidth of 25GB/s. Therefore, the maximum number of GPUs that can be saturated is 25GB/s / 4GB/s per GPU = 6.25 GPUs. Since you can't have a fraction of a GPU, the answer is 6 GPUs.

NEW QUESTION # 49
Consider a scenario where you are running a CUDA application on an NVIDIA GPU. The application compiles successfully but crashes during runtime with a *CUDA ERROR ILLEGAL ADDRESS* error. You've carefully reviewed your code and can't find any obvious out- of-bounds memory accesses. What advanced debugging techniques could help you pinpoint the source of this error?

• A. Employ the CUDA Debugger (cuda-gdb) to step through the code and inspect variable values and memory contents.
• B. Utilize NVIDIA Nsight Systems to profile the application and identify memory allocation patterns.
• C. Enable ECC (Error Correction Code) memory on the GPU to detect and correct memory errors.
• D. Reduce the block size used in CUDA kernels to decrease the likelihood of shared memory conflicts.
• E. Use 'cuda-memcheck' to detect memory access errors at runtime.

Answer: A,B,E

Explanation:
'cuda-memcheck' (A) is specifically designed to detect memory access errors. 'cuda-gdb' (B) allows for detailed code inspection. NVIDIA Nsight Systems (C) provides profiling information that can help identify memory allocation issues that lead to the error. Enabling ECC (D) might mask the symptom but doesn't fix the underlying code error, also ECC is enabled by default, so that's a red herring. Reducing block size (E) is only relevant if shared memory is involved and potentially overflowing.

NEW QUESTION # 50
After successfully installing the NVIDIA Container Toolkit and configuring the Docker runtime, you attempt to run a container that requires GPU access. However, the container fails to start with an error indicating that no GPUs are detected. You've verified that 'nvidia-smi' works on the host. Which of the following could be potential causes for this issue? (Select all that apply)

• A. The NVIDIA Container Toolkit package is corrupted and needs to be reinstalled.
• B. The Docker daemon was not restarted after configuring the NVIDIA runtime.
• C. The container image itself is missing the necessary CUDA libraries.
• D. The NVIDIA drivers are not compatible with the kernel version running on the host.
• E. The '-gpus all' flag was not included when running the 'docker run' command.

Answer: B,D,E

Explanation:
Several factors can cause this issue. A driver-kernel incompatibility (A) prevents the NVIDIA drivers from properly communicating with the hardware. Forgetting to restart the Docker daemon (B) means the configuration changes applied by 'nvidia-ctk' are not active. The '-gpus all' (or equivalent) flag (C) is mandatory to explicitly request GPU resources for the container. Corrupted toolkit (D) would likely present installation failures earlier. Missing CUDA libraries (E) would likely lead to runtime errors within the container, not a failure to detect the GPUs in the first place.

NEW QUESTION # 51
A user wants to restrict a Docker container to use only GPUs 0 and 2. Which command achieves this?

• A. docker run -e NVIDIA_VISIBLE_DEVICES=0,2 nvidia/cuda:12.1-base nvidia-smi
• B. docker run --device /dev/nvidia0,/dev/nvidia2 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 --gpus '"device=0,2"' nvidia/cuda:12.1-base nvidia-smi

Answer: D

Explanation:
With the advent of the NVIDIA Container Toolkit and modern Docker versions (19.03+), the --gpus flag is the official, verified method for resource allocation. To restrict a container to specific hardware IDs, the syntax requires a specific string format: --gpus '"device=0,2"'. This tells the NVIDIA Container Runtime to map only those specific physical GPU devices into the container's namespace. While environment variables like NVIDIA_VISIBLE_DEVICES (Option B) were used in older "nvidia-docker2" setups, they are now considered legacy and can be overridden by the more modern --gpus flag. Option D is incorrect because simply mapping the device nodes (/dev/nvidiaX) is insufficient; the container also needs the appropriate volume mounts for the NVIDIA drivers and libraries, which the --gpus flag handles automatically. This precise isolation is critical in multi-tenant AI environments to ensure that a single developer or job doesn't accidentally utilize the entire 8-GPU tray of a DGX H100.

NEW QUESTION # 52
......

Practice NCP-AII Questions With Certification guide Q&A from Training Expert 2Pass4sure: https://endexam.2pass4sure.com/NVIDIA-Certified-Professional/NCP-AII-actual-exam-braindumps.html