GPU Networks Management and Operations

A GPU network is a managed network object used to interconnect multiple GPUs within a single project on a high-bandwidth, low-latency network (Nvidia Spectrum-X East-West dedicated GPU network). GPU Networks provide tenancy isolation for such GPU-to-GPU dedicated inteconnectivity.

A GPU network has an ID, Name and VXLAN Network Identifier (VNI). In Nvidia Spectrum-X architecture, the CIDR is common to the whole fabric. A GPU network’s lifecycle is managed as follows: System-level by admins, account-level by tenant admins, and project-level by members with permissions to the project.

The GPU Networks Management page shows the GPU networks that are available to the account and their current assignment state.

The list view includes these fields:

• Name

  The display name of the GPU network. Use it to identify the network in the list and in action dialogs.

• VNI Number

  The VNI value (VXLAN) for the GPU network. Use it to identify the network segment when you create a new GPU network.

• State

  The current state of the GPU network. Use it to confirm whether the network is active or in another status.

• CIDR

  The address range for the GPU network. Use it to see the network scope.

• Project

  The project currently associated with the GPU network, when one is assigned.

• Account

  The account currently associated with the GPU network.

• ID

  The system ID of the GPU network. Use it when you need a unique identifier.

• Created At

  The date when the GPU network entry was created.

• Updated At

  The date when the GPU network entry was last updated.

The page supports these lifecycle operations:

• Add a GPU network

• Assign a GPU network

• Release a GPU network from an account

• Release a GPU network from a project

• Delete a GPU network

• Force delete a GPU network

• Delete multiple GPU networks

GPU Networks Management

Viewing GPU networks

To see the GPU networks that are available to the account:

1. In the left navigation pane, go to Account Networking.

2. Select GPU Networks Management.

   The page opens in List view.

3. Review the GPU networks table.

4. Use the table columns to inspect each GPU network.

   The list view includes these fields:

   • Name

     The display name of the GPU network. Use it to identify the GPU network in the list.

   • VNI Number

     The numeric VNI value (VXLAN) for the GPU network. Use it to identify the GPU network record.

   • State

     The current state of the GPU network. Use it to check its current condition.

   • CIDR

     The network address range for the GPU network.

   • Project

     The project associated with the GPU network, if a project assignment exists.

   • Account

     The account associated with the GPU network, if an account assignment exists.

   • ID

     The system ID of the GPU network.

   • Created At

     The date and time when the GPU network was created.

   • Updated At

     The date and time when the GPU network was last updated.

After selecting a GPU network, the UI exposes actions for that item, based on the current state and role permissions:

• Assign

• Release from Account

• Release from Project

• Delete

• Force Delete

Adding a GPU network

Use this action to create a new GPU network entry.

1. Open Account Networking > GPU Networks Management.

   The list of GPU networks appears.

2. Click Add.

   The Add GPU Network dialog opens.

   Enter values for these fields:

   • Name

     A descriptive name for the new GPU network.

     Required.

   • VNI Number

     The VNI value (VXLAN) for the new GPU network.

     Required.

   • Account

     The account pool to assign after creation.

     Use this field to place the new GPU network in an account pool.

   • Project

     The project pool to assign after creation.

     Use this field to place the new GPU network in a project pool.

3. To save the configuration, select Ok.

Assign a GPU network

Use this action to place an existing GPU network into an account pool or project pool.

1. Open Account Networking > GPU Networks Management.

   The list of GPU networks appears.

2. Select the GPU network to assign to an account pool or project pool.

3. Select Assign.

   The Assign GPU Network dialog opens.

   In the dialog, review or set these fields:

   • Account

     The account pool for the GPU network. When the GPU network is already assigned to an account pool, this field is shown with the current value and is not editable.

   • Project

     The project pool for the GPU network. The available project pools follow the selected account pool when an account is set.

4. To save the configuration, select Ok.

Release a GPU network from an account

Use this action to remove an account-level assignment from a GPU network.

1. Open Account Networking > GPU Networks Management.

   The list of GPU networks appears.

2. Select the GPU network to release from an account.

3. Select Release from Account.

   The Release from Account dialog opens.

   In the confirmation dialog, review these values:

   • Name

     The GPU network that will be released.

   • ID

     The system ID of the GPU network.

   • Account

     The account from which the GPU network will be released.

4. To confirm releasing the selected GPU network, select Ok.

Release a GPU network from a project

Use this action to remove a project-level assignment from a GPU network.

1. Open Account Networking > GPU Networks Management.

   The list of GPU networks appears.

2. Select the GPU network to release from a project.

3. Select Release from Project.

   The Release from Project dialog opens.

   In the confirmation dialog, review these values:

   • ID

     The system ID of the GPU network.

   • Project

     The project from which the GPU network will be released.

4. To confirm releasing the selected GPU network, select Ok.

Delete a GPU network

Use this action to remove one GPU network entry.

1. Open Account Networking > GPU Networks Management.

   The list of GPU networks appears.

2. Select the GPU network to delete.

3. Select Delete.

   The Delete GPU Network dialog opens.

   In the confirmation dialog, review these values:

   • Name

     The GPU network that will be deleted.

   • ID

     The system ID of the GPU network.

4. To confirm deleting the selected GPU network, select Ok.

Force delete a GPU network

Use this action when a standard delete does not complete.

1. Open Account Networking > GPU Networks Management.

   The list of GPU networks appears.

2. Select the GPU network to delete.

3. Select Force Delete.

   The Force Delete GPU Network dialog opens.

   In the confirmation dialog, review these values:

   • Name

     The GPU network that will be deleted.

   • ID

     The system ID of the GPU network.

4. To confirm force deleting the selected GPU network, select Ok.

Delete multiple GPU networks

Use this action to remove more than one GPU network at the same time.

1. Open Account Networking > GPU Networks Management.

   The list of GPU networks appears.

2. Select the GPU networks to delete.

3. Select Delete.

   The Delete GPU Network dialog opens, displaying the list of selected GPU network names that will be deleted.

4. To confirm deleting the selected GPU networks, select Ok.

Recommended best practices

• Use clear, unique names so admins can identify GPU networks quickly.

• Record the VNI (VXLAN) number before you create the GPU network.

• Review the Account and Project columns before you release or delete a GPU network.

• Use standard delete first. Use force delete only when required.

• Review all selected entries before you run a batch delete.

Troubleshooting

• Add is not available.

  The add action is available to system admins.

• Assign does not let you change the account pool.

  When a GPU network already has an account pool assignment, the Account field is shown as disabled in the assign dialog.

• Release from Account is not shown.

  This action is hidden when the GPU network is not assigned to an account pool, or when it is already assigned to a project pool.

• Release from Project is disabled.

  This action is disabled when the GPU network is not assigned to a project pool.

• Batch delete is unavailable.

  Batch delete is disabled when any selected GPU network is assigned to an account pool.

• The delete action fails.

  Retry the delete. If the standard delete does not complete, review whether Force Delete is appropriate.