NCP-AII Question 11
Single answerYou are managing an AI training cluster with several NVIDIA GPUs. One of the training jobs reports inconsistent performance, and you suspect a hardware issue. Using NVIDIA System Management Interface (nvidia-smi), you notice that one GPU consistently shows high error counts for ECC (Error Correcting Code) and thermal throttling. What is the most likely faulty component, and what should you do next?
- A
The GPU is faulty; replace the GPU immediately.
- B
The power supply is faulty; replace the power supply unit.
- C
The GPU cooling system is insufficient; check and clean the GPU fans and thermal paste.
- D
The motherboard PCIe slot is faulty; reseat the GPU or change the slot.
Show answer and explanation
Correct answer: C
Explanation
High ECC error counts and thermal throttling often indicate that the GPU is overheating rather than defective. Cleaning the cooling system, ensuring proper airflow, or replacing thermal paste is a logical first step to address these symptoms. Replacing the GPU or other hardware components without diagnosing the root cause could lead to unnecessary downtime and costs.
- A. Incorrect.
While the GPU might appear faulty, high ECC error counts and thermal throttling are often symptoms caused by overheating rather than a hardware defect. Replacing the GPU without further investigation would be premature.
- B. Incorrect.
Power supply issues typically result in system instability or shutdowns, not specific GPU errors or thermal throttling.
- C. Correct.
High ECC errors and thermal throttling are common signs of overheating. Cleaning the GPU fans, ensuring proper airflow, or replacing thermal paste can resolve these issues without replacing hardware.
- D. Incorrect.
A faulty PCIe slot might cause connectivity issues or prevent the GPU from being recognized, but it does not typically cause high ECC errors or thermal throttling.