NCP-AII Question 41
Select 3You are tasked with setting up an NVIDIA NVLink™ Switch in an AI compute cluster to enable high-speed interconnects between GPUs. During the configuration, you need to ensure proper hardware and software setup for optimal performance. Which steps should you take to correctly install and configure the NVLink Switch?
- A
Ensure that the GPUs in the nodes are NVLink-compatible and support NVSwitch.
- B
Update the system firmware and drivers to the latest NVIDIA-supported versions.
- C
Connect the NVLink Switch to a standard Ethernet network for compatibility.
- D
Verify the topology configuration in the NVIDIA System Management Interface (nvidia-smi).
- E
Disable ECC (Error Correcting Code) memory to maximize bandwidth for NVLink communication.
Show answer and explanation
Correct answers: A, B, D
Explanation
To properly install and configure an NVIDIA NVLink Switch, you must ensure hardware compatibility (NVLink-compatible GPUs), update firmware and drivers for compatibility and performance, and verify the topology using tools like nvidia-smi. Connecting the NVLink Switch to Ethernet or disabling ECC memory are not required steps and could lead to suboptimal performance or system instability.
- A. Correct.
Correct. NVLink and NVSwitch require GPUs that support these technologies. Ensuring compatibility is essential for the setup to work.
- B. Correct.
Correct. Updating firmware and drivers ensures that the system can fully utilize NVLink and NVSwitch features and avoids potential compatibility issues.
- C. Incorrect.
Incorrect. NVLink Switch is not connected to standard Ethernet networks; it is a dedicated high-speed interconnect for GPUs.
- D. Correct.
Correct. Verifying the topology in NVIDIA System Management Interface (nvidia-smi) is a critical step to ensure proper connectivity and configuration.
- E. Incorrect.
Incorrect. Disabling ECC is not a requirement for NVLink communication; ECC ensures memory integrity and is recommended for AI workloads.