NCP-AII Question 82
Select 3An organization is preparing to deploy an AI workload on an NVIDIA GPU-enabled server and wants to validate that the hardware is functioning correctly. Which of the following steps should be performed to ensure the hardware can handle the intended workload effectively?
- A
Run NVIDIA System Management Interface (nvidia-smi) to verify GPU visibility and utilization.
- B
Perform a stress test using an AI workload or synthetic benchmark like NVIDIA TensorRT Inference Server.
- C
Check the server's physical network connections to ensure there is no packet loss.
- D
Verify the GPU driver and CUDA toolkit versions are compatible with the workload's requirements.
- E
Run the operating system's default system diagnostics for CPU and memory to validate hardware.
Show answer and explanation
Correct answers: A, B, D
Explanation
To validate hardware operation for AI workloads, it is essential to ensure the GPUs are detected and operational (nvidia-smi), capable of handling the workload (stress testing or benchmarking), and configured with compatible drivers and toolkits. These steps ensure the hardware is ready and optimized for the intended AI workload.
- A. Correct.
Running
nvidia-smiis a critical step to verify that the GPUs are properly detected and operational, as well as to check their current utilization and health metrics. - B. Correct.
Performing a stress test or running a synthetic benchmark ensures that the GPUs can handle the expected workload under load conditions and are functioning as intended.
- C. Incorrect.
Checking network connectivity is important for distributed training setups but is not directly related to validating GPU hardware operation for workloads.
- D. Correct.
Verifying that the GPU driver and CUDA toolkit versions match the workload's requirements is essential, as incompatible versions can cause errors or degraded performance.
- E. Incorrect.
Running the operating system's default CPU and memory diagnostics is unrelated to validating GPU hardware. Specialized tools like
nvidia-smiare required for GPU-specific validation.