NCA-AIIO Question 109
Select 4A data center administrator is tasked with monitoring the GPUs in their AI infrastructure to ensure optimal performance and prevent system downtime. Which of the following key measures should the administrator monitor to achieve this goal?
- A
GPU utilization percentage
- B
Memory utilization of the GPU
- C
CPU clock speed
- D
GPU temperature
- E
Disk I/O operations
- F
Error rates related to GPU tasks
Show answer and explanation
Correct answers: A, B, D, F
Explanation
Monitoring GPUs in an AI infrastructure requires focusing on measures directly related to GPU health and performance, such as utilization, memory usage, temperature, and error rates. These metrics help administrators ensure that GPUs are operating efficiently and within safe operational limits, while also identifying potential issues early. Metrics like CPU clock speed and disk I/O, while important for overall system performance, are not specific to GPU monitoring.
- A. Correct.
Monitoring GPU utilization percentage helps identify whether the GPUs are underutilized or overloaded, which is critical for performance optimization.
- B. Correct.
Memory utilization of the GPU provides insights into whether the available memory is sufficient for workloads, preventing memory bottlenecks.
- C. Incorrect.
While CPU clock speed is important for general system performance, it is not a direct measure related to GPU monitoring in an AI infrastructure context.
- D. Correct.
GPU temperature is a key measure to ensure the hardware operates within safe thermal limits, preventing overheating and hardware damage.
- E. Incorrect.
Disk I/O operations are not directly related to GPU performance but are more relevant to storage monitoring.
- F. Correct.
Error rates related to GPU tasks, such as ECC (Error-Correcting Code) errors, can indicate potential hardware issues or workload-specific problems that need to be addressed.