Docs

Elastic Compute Service

GPU instances issue

Solution for nvidia-fabricmanager.service status exception

Solution for nvidia-fabricmanager.service status exception

This topic mainly describes how to troubleshoot the "nvidia-fabricmanager.service" exception when you use a GPU instance.

Procedure

Log in to the target instance. For operation details, refer to Logging in to Linux instances.

Execute the following command to confirm whether the nvidia-fabricmanager service status is active (running).

systemctl status nvidia-fabricmanager.service

A sample of normal output is as follows:

If the service is not in active (running) status, perform the following operations:

Confirm whether the instance needs the nvidia-fabricmanager.service service.

Only for HPC-optimized instances, the nvidia-fabricmanager.service service is needed. In instances of other types, it is considered normal for the nvidia-fabricmanager service to have an abnormal status or to be absent. Learn more about HPC-optimized instances.

Confirm whether the version of nvidia-fabricmanager.service service matches that of the instance driver.

Execute the following command to view the service version.

/usr/bin/nv-fabricmanager -v

Execute the following command to view the driver version.

nvidia-smi

Check whether the versions of the two match each other.

To start the service successfully, make sure the two versions match.

If the versions match, execute the following command to restart and then check the service status.

systemctl restart nvidia-fabricmanager.service

To obtain support when the method described in this topic does not resolve the exception, submit a ticket or contact your account manager.

Last updated: 2026.06.18 17:49:51