NCP-AII Question 20
Select 3You are managing an NVIDIA DGX system and notice that one of the GPUs has failed. The system needs to remain operational during the replacement process. Which steps should you take to safely replace the faulty GPU while minimizing downtime?
- A
Check the system logs to confirm the GPU failure and identify the faulty GPU.
- B
Power off the entire system before replacing the GPU to ensure safety.
- C
Use the NVIDIA System Management Interface (nvidia-smi) to isolate the faulty GPU.
- D
Ensure proper grounding and use ESD (Electrostatic Discharge) protection before handling the GPU.
- E
Hot-swap the GPU without shutting down the system, as modern GPUs support this feature.
Show answer and explanation
Correct answers: A, C, D
Explanation
To safely replace a faulty GPU in an NVIDIA DGX system while minimizing downtime, you must first identify the issue using system logs. Isolating the GPU with tools like nvidia-smi ensures that the system doesn't interact with the faulty component during the replacement process. Additionally, grounding and ESD protection are essential to protect the hardware. Powering off the entire system or attempting hot-swapping are not suitable actions in most cases, as they either increase downtime or risk damage.
- A. Correct.
Checking system logs is a critical first step to confirm the GPU failure and identify the faulty component precisely. This ensures you are addressing the correct issue.
- B. Incorrect.
Powering off the entire system is not always necessary, especially in systems designed for hot-swappable components. Shutting down the whole system would result in unnecessary downtime.
- C. Correct.
Using nvidia-smi to isolate the GPU is a best practice to prevent the system from attempting to access the faulty GPU during the replacement process.
- D. Correct.
Proper grounding and ESD protection are mandatory to avoid damaging delicate components, including the GPU, during handling.
- E. Incorrect.
Hot-swapping GPUs is not typically supported in NVIDIA systems. Attempting this could result in hardware damage or system instability.