NCP-AII Question 104
Select 3You are managing an NVIDIA DGX-based AI cluster in a data center. During a routine inspection, you notice inconsistent GPU performance metrics across nodes. Upon investigation, you discover that the airflow in the server racks is obstructed, and the temperature in the rack exceeds recommended levels. Which of the following steps should you take to address the issue and ensure optimal physical layer management?
- A
Reorganize cable management to improve airflow and reduce obstructions.
- B
Ensure proper spacing between server nodes within the rack.
- C
Increase the fan speed of GPU cards via software tools.
- D
Verify that the data center's cooling systems are functioning correctly.
- E
Replace the GPUs as they may have been permanently damaged.
Show answer and explanation
Correct answers: A, B, D
Explanation
Physical layer management in an AI infrastructure involves ensuring proper airflow, cooling, and hardware placement to prevent performance degradation. In this scenario, obstructed airflow and high temperatures must be addressed by reorganizing cables, spacing equipment properly, and verifying cooling system functionality. These steps ensure that the thermal environment supports optimal GPU performance without unnecessary hardware replacement.
- A. Correct.
Proper cable management is critical for maintaining unobstructed airflow in server racks, which directly impacts cooling efficiency.
- B. Correct.
Adequate spacing between servers ensures that heat can dissipate more effectively, reducing the risk of overheating.
- C. Incorrect.
While increasing GPU fan speed might provide a temporary solution, it does not address the root cause of the airflow obstruction.
- D. Correct.
Ensuring the cooling systems in the data center are operating correctly is essential to maintaining consistent temperature levels and preventing thermal issues.
- E. Incorrect.
Although overheating can damage GPUs, this step is premature unless specific diagnostics confirm permanent damage.