Documentation clarification: ConnectX-7 dual-port recommendation for the NVQLink latency demo

Hi CUDA-Q team,

Thank you for releasing CUDA-Q Realtime and the NVQLink latency demo.

We are currently preparing the hardware to reproduce the NVQLink latency demo, and we have a question regarding the recommended NIC configuration.

The documentation states:

"For this experiment, we recommend using NVIDIA ConnectX-7 NIC with dual QSFP ports. Prior ConnectX generations may not have all the capabilities required."

However, it is not clear to us why a dual-port ConnectX-7 is recommended.

Question 1

In several reference photos of the lab setup, the RFSoC board appears to be connected to the host using only a single QSFP cable.

Does the latency demo actually use only one physical port on the ConnectX-7, or are both ports required during execution?

If only one port is used, could you clarify why the documentation specifically recommends a dual-port ConnectX-7 NIC?

Question 2

We are deciding between the following ConnectX-7 adapters:

• ConnectX-7, single-port 400GbE (OSFP, MCX75310AAS-NEAT)
• ConnectX-7, dual-port 200GbE (QSFP112, MCX755106AS-HEAT)

Would a single-port 400G ConnectX-7 also work for reproducing the latency demo, or is the dual-port QSFP112 model strongly recommended?

If the recommendation is intentional, could you briefly explain the reason (for example, hardware compatibility, supported topology, software implementation, or validation platform)?

This clarification would greatly help users who are purchasing hardware to reproduce the demo.

Thank you!

Start with the quoted ConnectX-7 recommendation, the NVQLink latency demo description, and the reference hardware photos mentioned in the issue. Confirm whether execution uses one or both QSFP ports and whether the single-port 400GbE model is supported, then update the documentation with the validated recommendation and its rationale.