After launched the 8 gpu tdx vm, nvswitch raise error

Accelerated Computing Confidential Computing
1

After loaded the vm , nvidia-fabricmanager raise following error

[Jan 06 2026 18:33:42] [INFO] [tid 1792] Fabric Manager version 580.95.05 is running with the following configuration options
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Logging level = 4
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Logging file name/path = /var/log/fabricmanager.log
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Append to log file = 1
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Max Log file size = 1024 (MBs)
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Use Syslog file = 0
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Fabric Manager communication ports = 16000
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Fabric Mode = 0
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Fabric Mode Restart = 0
[Jan 06 2026 18:33:42] [INFO] [tid 1792] FM Library communication bind interface = 127.0.0.1
[Jan 06 2026 18:33:42] [INFO] [tid 1792] FM Library communication unix domain socket = 
[Jan 06 2026 18:33:42] [INFO] [tid 1792] FM Library communication port number = 6666
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Continue to run when facing failures = 0
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Option when facing GPU to NVSwitch NVLink failure = 0
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Option when facing NVSwitch to NVSwitch NVLink failure = 0
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Option when facing NVSwitch failure = 0
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Abort CUDA jobs when FM exits = 1
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Fabric Manager - Subnet Manager IPC socket = unix:/var/run/nvidia-fabricmanager/fm_sm_ipc.socket
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Fabric Manager - Subnet Manager management port GUID = 
[Jan 06 2026 18:33:42] [INFO] [tid 1792] Disabling RPC mode for single node configuration.
[Jan 06 2026 18:33:43] [ERROR] [tid 1792] request to query NVSwitch device information from NVSwitch driver failed with error:WARNING Nothing to do [NV_WARN_NOTHING_TO_DO]

in the dmesg, it raise following error

[   20.049438] nvidia-nvswitch: Probing device 0000:09:00.0, Vendor Id = 0x10de, Device Id = 0x22a3, Class = 0x68000 
[   20.087400] nvidia-nvswitch0: Failed to initialize device : -8
[   20.090064] nvidia-nvswitch0: Failed to initialize device : -19
[   20.092193] nvidia-nvswitch: Probing device 0000:0a:00.0, Vendor Id = 0x10de, Device Id = 0x22a3, Class = 0x68000 
[   20.127424] nvidia-nvswitch0: Failed to initialize device : -8
[   20.130935] nvidia-nvswitch0: Failed to initialize device : -19
[   20.137067] nvidia-nvswitch: Probing device 0000:0b:00.0, Vendor Id = 0x10de, Device Id = 0x22a3, Class = 0x68000 
[   20.171460] nvidia-nvswitch0: Failed to initialize device : -8
[   20.175686] nvidia-nvswitch0: Failed to initialize device : -19
[   20.182631] nvidia-nvswitch: Probing device 0000:0c:00.0, Vendor Id = 0x10de, Device Id = 0x22a3, Class = 0x68000 
[   20.215454] nvidia-nvswitch0: Failed to initialize device : -8
[   20.219718] nvidia-nvswitch0: Failed to initialize device : -19

I already

  1. configure the 8xgpus + 4xswitch to ppcie=on and cc=off
  2. tried several times hard reboot.
  3. change several servers from different vendor

but the issue can not fixed still.

2

@zjeff is this Hopper PPCIe or Blackwell MPT? Use of fabric manager in host / guest is different based on GPU generation.