# GPU network switches

GPU network switches are high-performance networking platforms designed to
accelerate AI, machine learning, and GPU computing workloads.
They provide ultra-low latency, high-bandwidth per port, and in-network
computing features to connect thousands of GPUs, preventing bottlenecks in
data-intensive environments.

The zCompute GPU Network comprises NVIDIA Ethernet switches configured in a
Spectrum-X Architecture.
The GPU Network switches defined in zCompute are the leaf switches directly
attached to the GPU’s dedicated NICs (SuperNICs).

## GPU network switches management

The GPU Network Switches page stores GPU network switch records in a
central place.

You can use this page to review existing records, add a new record,
update a record, or remove one.

### Viewing GPU network switches

To view GPU network switches:

1. Go to **Region Networking > GPU Network Switches**.

   The GPU network switches list view displays.

   This view shows the current GPU network switch records with the the
   default columns:
   * **Name**

     The switch record name.

     Use it to identify the record in the list, in success or failure
     messages, and in the delete dialog.
   * **Status**

     The current state that the UI reports for the switch record.

     Use it for a quick state check from the list page.
   * **Hostname**

     The host address saved for the switch record.

     Use it to confirm that the record points to the intended switch host.
   * **API Port**

     The API TCP port saved for the switch record.

     Use it to confirm the port value used by the record.
   * **Username**

     The saved user name for the switch record.

     Use it to confirm which account name the record is set to use.
   * **Rail Group**

     The rail group value saved for the switch record.

     Use it to review the value that was set during creation.
   * **SU ID**

     The scalable unit value saved for the switch record.

     Use it to review the value that was set during creation.
   * **Router Distinguisher**

     The router distinguisher value saved for the switch record.

     Use it to verify the saved distinguisher from the list page.
   * **Verify SSL**

     Indicates whether SSL verification is enabled for the switch record.

     Use it to confirm the setting that was chosen during creation.

   You can also use the filter selection to display these optional columns:
   * **ID**

     The record ID.

     Use it when you need the unique record value.
   * **Created At**

     The record creation date and time.

     Use it for audit and tracking work.
   * **Updated At**

     The record’s last update date and time.

     Use it for audit and tracking work.

When no records exist, the page shows a message that no GPU network
switches are defined.

#### Viewing GPU network switch details

To view a GPU network switch details:

1. Go to **Region Networking > GPU Network Switches**.

   The GPU network switches list view displays.
2. Select a GPU network switch to view its details.

   The GPU network switch details page opens.

### Adding a GPU network switch

To add a GPU network switch:

1. Open **Region Networking > GPU Network Switches**.

   The GPU network switches list view displays.
2. Select **+ Add**.

   The **Add GPU Network Switch** dialog opens.
3. Enter new GPU network switch values:
   * **Name**

     The GPU network switch display name.

     Required.
   * **Hostname**

     The host address.

     Required.
   * **API TCP Port**

     The API port for the GPU network switch.

     Valid values: **1** to **65536**.

     Default: **8765**
   * **Verify SSL**

     Toggle to enable/disable SSL verification for the GPU network switch.

     Default: **Enabled**.
   * **Rail Group**

     Represents a collection of network switches that connect the
     same-index GPU from every server across multiple racks.
     For example, Rail Group 0 connects all GPU 0s,
     Rail Group 1 connects all GPU 1s, and so on.

     Valid values: 0 or higher.

     Required.
   * **Router Distinguisher**

     Used to create unique addresses from overlapping IP address spaces,
     ensuring they can be routed correctly across a virtualized network.

     Required.
   * **Scalable Unit**

     ID of a set of resources for rapid scaling purposes.

     Instead of building a node-by-node, organizations can
     use SUs to deploy a standardized set of resources simultaneously.
     This modularity allows for rapid deployment and ensures the
     resulting cluster is balanced for maximum performance.

     Valid values: **0** to **31**.

     Required.
   * **Username**

     The username for GPU network switch.

     Required.
   * **Password**

     The password for GPU network switch username.

     Required.
4. To save the configuration, select **Finish**.

### Modify a GPU network switch

Update a switch record when connection details or credentials change:

1. Open **Region Networking > GPU Network Switches**.

   The GPU network switches list view displays.
2. Select a GPU network switch to modify.

   The GPU network switch details page opens.
3. Select  **Modify**.

   The **Modify GPU Network Switch** dialog opens.
4. The following fields can be updated:
   * **Name**

     The GPU network switch display name.
   * **Hostname**

     The host address.
   * **API TCP Port**

     The API port for the GPU network switch.
   * **Router Distinguisher**

     IP address used to create unique addresses from overlapping IP
     address spaces.
   * **Username**

   Required. Use it to change the saved user name.
   * **Change Password**
     1. Select the checkbox to enable changing the password for GPU
        network switch username.

        The **Password** prompt appears.
     2. Enter a new password for the GPU network switch username.
5. To save the GPU network switch updates, select **Finish**.

### Delete a GPU network switch

To remove a GPU network switch record when it is no longer needed.

1. Open **Region Networking > GPU Network Switches**.

   The GPU network switches list view displays.
2. Select a GPU network switch to delete.

   The GPU network switch details page opens.
3. Select  **Delete**.

   The **Delete GPU Network Switch** dialog opens, displaying the name of
   the GPU network switch selected for deletion.
4. To confirm deletion of the selected GPU network switch, select
   **Delete**.

## Recommended best practices

* Use a clear naming standard.

  The name appears in the list, in action messages, and in the delete flow.
* Check the hostname before you save the record.

  It is required and stays visible in the main table.
* Leave the API TCP port at the default only when that matches your switch
  configuration.

  The default value is **8765**.
* Set the **Verify SSL**, **Rail Group** and **Scalable Unit** values
  carefully during creation.

  The current UI displays these fields only in the **+ Add** dialog, and
  not when modifying or deleting.
* Review the router distinguisher value before saving a new entry or
  update.

  It is required in both **+ Add** and **Modify** dialogs.
* Change the password only when needed.

  In the **Modify** dialog, a new password key and value pair is setting
  with the update payload only when **Change Password** is selected.
* Use the optional columns during audits.

  **ID**, **Created At**, and **Updated At** can help with tracking.

## Troubleshooting

### The GPU Network Switches page does not appear in the menu

GPU Network Switches is part of the administrative Region Networking
menu.

* Confirm that you are using an account with system admin access.
* Open **Region Networking** and check the left navigation again.

### The GPU Network Switches list is empty

An empty list can be normal when no switch records exist.

* Check for the message that no GPU network switches are defined.
* Add a switch if this is a new setup.
* Refresh the page if you expect existing data.

### The form does not let you finish

The dialog prevents completion when required fields are missing or invalid.

* Check **Name**.
* Check **Hostname**.
* Check **Router Distinguisher**.
* Check **Username**.
* Check **Password** in the **+ Add** dialog.
* Check **Rail Group** and **Scalable Unit** in the **+ Add** dialog
* Check the numeric range for **API TCP Port**.
* Check the numeric range for **Scalable Unit**.

### You cannot change verify SSL, Rail Group, or Scalable Unit

This behavior is expected in the current UI.

The  **Verify SSL**, **Rail Group**, and **Scalable Unit** fields are
available only in the **+ Add** dialog.

Plan these values before you create the record.

### Password is not updated

The modify dialog sends a password only when **Change Password** is
selected.

* Reopen **Modify**.
* Select **Change Password**.
* Enter the new password.
* Click **Finish**.

### A create, modify, or delete action fails

The UI defines success and failure messages for **+ Add**, **Modify** and
**Delete** actions.

* Read the error message.
* Recheck the values you entered.
* Verify that you selected the intended switch.
* Retry the action after you correct the input.