NCP-AII Question 46
Select 3You are setting up the network fabric ports for an AI infrastructure deployment using NVIDIA networking solutions. The environment consists of multiple GPU servers connected via NVIDIA Spectrum switches. Which actions should you take to ensure optimal configuration of the network fabric ports for the hosts?
- A
Enable RDMA over Converged Ethernet (RoCE) on the ports.
- B
Configure switch ports to operate in Layer 2 mode exclusively.
- C
Set the MTU (Maximum Transmission Unit) size to support jumbo frames, such as 9000 bytes.
- D
Enable port mirroring on all fabric ports for monitoring purposes.
- E
Verify that Priority Flow Control (PFC) is enabled for lossless traffic.
Show answer and explanation
Correct answers: A, C, E
Explanation
To set up network fabric ports for AI infrastructure, enabling RDMA over Converged Ethernet (RoCE), configuring the MTU for jumbo frames, and ensuring lossless traffic through Priority Flow Control (PFC) are essential steps. These configurations optimize the network for high-performance AI workloads, ensuring low latency and high throughput, which are critical requirements in distributed GPU environments.
- A. Correct.
Enabling RDMA over Converged Ethernet (RoCE) is critical for achieving high-performance, low-latency communication between GPU servers in AI workloads. This is a key recommendation for NVIDIA AI infrastructure.
- B. Incorrect.
While Layer 2 mode might work for specific setups, AI workloads in distributed environments often benefit from Layer 3 configurations for scalability and routing flexibility. Limiting the configuration to Layer 2 mode is not optimal in most AI applications.
- C. Correct.
Setting the MTU size to support jumbo frames is necessary to optimize data transmission efficiency and reduce overhead, which is particularly important in high-throughput AI workloads.
- D. Incorrect.
Port mirroring is typically used for troubleshooting and monitoring purposes but can introduce performance overhead. It is not required for normal operation of the network fabric in AI workloads.
- E. Correct.
Priority Flow Control (PFC) ensures lossless traffic for RDMA communications, which is critical for maintaining low latency and high performance in AI infrastructure networks.