---
title: "Connect Multiple DGX Sparks for Distributed Workloads — Troubleshooting"
canonical: "https://build.nvidia.com/spark/connect-multiple-sparks/troubleshooting.md"
---

# Connection and NVIDIA Sync issues

| Symptom | Cause | Fix |
| --- | --- | --- |
| A device does not appear in NVIDIA Sync | The device is off, on another network, or cannot accept SSH connections | Turn on the device. Make sure it and the computer that runs NVIDIA Sync are on the same local network. Test direct SSH access. |
| The GB10 check fails | The selected system is not a DGX Spark or GB10 device | Remove the unsupported system from the cluster. Cluster Assistant supports only DGX Spark and GB10 devices. |
| The software check fails | A device has an old system release | Update each device. Cluster Assistant requires the April 2026 system release or later. |
| The password check fails | NVIDIA Sync cannot use `sudo` on a device | Select **Fix Now** and enter the device password. Check that the user has `sudo` access. |
| Cluster Assistant finds the wrong layout | A cable is loose, the layout is not supported, or the Spark-facing switch ports are not on the same Layer 2 network | Reseat each cable and compare the links with **Connect the Devices**. Do not mix direct and switch links. For a switch, check that every Spark-facing port is up and on the same Layer 2 network. Then run the layout check again. |
| A switch link is down or is not 200 Gbit/s | The cable is not supported, the switch port uses the wrong mode, or the two ends do not agree on link settings | Check the cable and port mode. A 400 Gbit/s port may need to be split into two 200 Gbit/s ports. If the link stays down, check auto-negotiation and FEC in the switch maker's guide. |
| The cluster connects but data transfer is slow | A link can report 200 Gbit/s while the switch sends traffic through its CPU or the link records errors | Check hardware-offload status, switch CPU use, and port error counters. Test traffic in both directions. If the issue remains, contact the switch maker or NVIDIA support. |
| SSH setup times out | One device took more than five minutes | Retry the SSH step in Cluster Assistant. |

For more Cluster Assistant help, see the [NVIDIA Sync troubleshooting guide](https://docs.nvidia.com/sync/latest/cluster-assistant.html#troubleshooting).

# Manual setup issues

| Symptom | Cause | Fix |
| --- | --- | --- |
| The manual pre-check fails | A management IP, SSH login, password, or `sudo` setting is wrong | Fix the value in `config/my-cluster.json`, test SSH to each management IP, and run `--pre-validate-only` again. |
| No ConnectX-7 interface is up | A cable or port is not active | Reseat the cables, confirm the layout, reboot the devices, and run `ibdev2netdev` again. |