My QSFP ports on my DGX Spark are not working

I have a DGX Spark and an HP ZGX Nano G1n AI station and I was trying to connect them with a Naddod Q112-400G-CU0-5 cable through their QSFP ports. After some troubleshooting, Mellanox (ConnectX) hardware isn’t being detected at the PCI level – lspci only shows a Realtek Ethernet controller and a MediaTek Wi‑Fi/BT chip. The ZGX did work and was able to find the ports. I did some more troubleshooting and Claude thinks this is the issue:

The dmesg tells you exactly what’s happening — no need to guess:

cx7-pcie-hotplug MTKP0001:00: Cable removal

The system has a cx7-pcie-hotplug driver specifically for the ConnectX-7. It initialized successfully, then immediately detected “Cable removal” — meaning the internal PCIe cable between the motherboard and the CX7 module is physically disconnected or unseated. The LnkSta: Width x0 confirms zero PCIe lanes are active to it.

This is not a driver issue. The hardware connection is broken.

So, at this point, do I take it apart or does that void my warranty?

Do not take it apart, it may void. your warranty. Instead, please run https://nvidia.custhelp.com/app/answers/detail/a_id/5767/~/nvidia-dgx-spark-field-diagnostics then DM me the resulting logs on next steps.

Have you tried disconnecting the power of both devices for 1 minute? Leave link connected.
I had issues as well the first time. After 1 minute without power. I was able to boot and connect.
Good luck.

This is part of the CX7 powersavings feature introduced in the latest update. When there is no cable connected, the card will power down and will not be visible in lspci.
After plugging in the cable you should see it start working

I have tried powering down the DGX spark only. I will try both at the same time.

The cable has been plugged in the entire time to the port closest to the power button the whole time. I did unplug it and plug it back in once. It seems firmly seated.

I think I DM’d them to you. Let me know if you see anything out of the ordinary.

Hi Eric,

Per the logs you shared, your hardware seems healthy. Can you share when this symptom started? did the CX7 device ever detect for you?

Hi Eric, please try disabling hotplug support by removing this file
sudo rm -f /etc/nvidia/cx7-hotplug-enabled
And then rebooting

This doesn’t seem to have helped. After this was complete, I tried both methods in the Spark Stacking Guide and neither of them found the connection.

I have only had the Spark for a week and the ZGX for a couple days. The first time that I tried to connect them, I wasn’t able to. So far, I have tried powering them both down for 1+ minutes and removing the line the other moderator suggested.

This looks like an RMA case. Please contact my colleagues at NVIDIA DGX Spark Support and reference this thread for context.

Note: We can only assist with RMA requests for the DGX Spark Founder’s Edition. For an HP ZGX RMA, please contact HP directly.