You can create, manage, and publish vGPU policies to organization VDCs that use flex allocation model.

NVIDIA GRID vGPU is a graphics acceleration technology from NVIDIA that you can use to share a single graphics processing unit (GPU) among multiple virtual desktops. When you use NVIDIA GRID cards, installed on an x86 host, in a desktop and application virtualization solution running on vSphere 6.x and later, you can render application graphics with superior performance compared to non-hardware-accelerated environments. This capability is useful for graphics-intensive use cases such as designers in a manufacturing setting, architects, engineering labs, higher education, oil and gas exploration, clinicians in a healthcare setting, and for power users and knowledge workers who need access to rich 2D and 3D graphical interfaces.

Under Infrastructure Resources, you can view the vGPU Profiles that VMware Cloud Director loads from the vCenter Server clusters with virtual graphics processing unit (vGPU) capabilities. Each profile represents a type of vGPU. You can use a vGPU profile to create a vGPU policy that tenants can use for their virtual machines. NVIDIA vGPU profiles determine how many fixed share resources can be allocated to each VM from the total available memory.

You create and manage vGPU policies at the global or at the provider level. You can publish individual policies to one or more organization VDCs that use flex allocation model.

View and Manage vGPU Profile Information

When the hosts in a cluster backing the provider VDC resource pool have vGPU capabilities, VMware Cloud Director loads the vGPU profile information from vCenter Server. You can view, edit, and delete the vGPU profile information.

You select a vGPU profile during the creation of a vGPU policy. When you publish a vGPU policy, the vGPU profile names and instructions of the profiles you added to the policy become visible to the tenants.

Procedure

  1. From the top navigation bar, under Resources, click Infrastructure Resources.
  2. In the left panel, select vGPU Profiles.
  3. Click the radio button next to a vGPU profile name, and click Edit.
  4. (Optional) Edit the tenant-facing name.
  5. (Optional) Edit the tenant-facing instructions.
    You can provide additional instructions to your tenants, for example, you can provide a link to the NVIDIA GRID vGPU release notes document.
  6. Click Save.
  7. To delete a vGPU profile, click the radio button next to a vGPU profile name, and click Delete.
    The option to delete a vGPU profile appears for the vGPU profiles which the system detects as no longer valid in the underlying infrastructure and which are not used by any vGPU policies.
  8. To view vGPU profile metrics for their use, click the name of the profile and under Usage metrics you can view details about where this profile is used.
    You can view the names of the VMs and vApps that use the vGPU profile, usage count, policy name, organization, and organization VDC information.

Create a vGPU Policy

Starting with VMware Cloud Director 10.3.2, to define the placement and sizing settings of VMs that require vGPU resources, you can create vGPU policies.

Prerequisites

Verify that at least one vCenter Server host has an NVIDIA graphics device attached and all required vSphere Installation Bundles (VIBs) are installed on the host.

Procedure

  1. From the top navigation bar, select Resources and click Cloud Resources.
  2. Navigate to the Create a vGPU Policy wizard.
    • In the left panel, select vGPU Policies, and click New.
      1. In the left panel, select Provider VDCs.
      2. Click the name of a provider VDC with the NVIDIA icon indicating vGPU capabilities.
      3. Under Policies, select vGPU, and click New.
  3. On the General page of the Create vGPU Policy wizard, enter a vGPU policy name, and optionally, a description.
  4. Click Next.
  5. Select a vGPU profile and PCI device count associated with this policy, and click Next.
    By selecting a count, you select the number of PCI devices that you can attach to a VM that you create by using this policy.
  6. Select which provider VDC clusters can have access to the policy, and click Next.
    If you navigated to the wizard through the Provider VDC tab, only the selected provider VDC is visible. Selecting No creates a global vGPU policy that all provider VDC clusters can access.
  7. If you want to define the placement of a VM on a host or group of hosts, select Yes and select one or more VM groups.

    A VM group has a direct affinity to a host group and represents the host group to which it has the affinity.

    You can select one VM group per cluster.

  8. Click Next.
  9. If you want to define the compute resource allocation for VMs, select Yes, and click Next.
    1. Select the CPU allocation settings that you want to apply to the policy, and click Next.
    2. Select the memory allocation settings that you want to apply to the policy, and click Next.
      Note: The memory reservation guarantee must always be 100%.
    3. Configure additional VM settings as extraConfig parameters, and click Next.
      The Extra Configurations VDC compute policy attribute is a mapping between key and value pairs applied as extra configuration values on a VM.
  10. Review the vGPU policy settings, and click Finish.

Add a vGPU Policy to an Organization VDC

When you create a vGPU policy, it is not visible to tenants. You can publish a vGPU policy to an organization VDC to make it available to tenants.

Publishing a vGPU policy to an organization VDC makes the policy visible to tenants. The tenant can select the policy when they create a new standalone VM or a VM from a template, edit a VM, add a VM to a vApp, and create a vApp from a vApp template. You cannot delete a vGPU policy that is available to tenants.

Prerequisites

  • Verify that you have at least one organization VDC in your environment. See Create an Organization Virtual Data Center.
  • Verify that you have at least one vGPU policy. See Create a vGPU Policy.
  • Verify that the organization VDC to which you want to publish a vGPU policy is using flex allocation model.
  • Verify that the organization VDC to which you want to publish a vGPU policy belongs to a provider VDC scoped in the vGPU policy. Alternatively, verify that the vGPU policy you want to publish is global and does not have a provider VDC scope.

Procedure

  1. From the top navigation bar, select Resources and click Cloud Resources.
  2. In the left panel, click Organization VDCs.
  3. Select an organization VDC and under Policies, select the vGPU tab.
  4. Click Add.
  5. Select the vGPU policies that you want to add to the organization VDC, and click OK.

What to do next

  • Select a policy and click Remove to unpublish the policy.
  • Select a vGPU policy and click Set as default to make that policy appear as the default choice for the tenants during a VM and vApp creation and VM edit. If there is more than one vGPU policy published for an organization VDC, the tenant can select a different policy from the default one.

Delete a vGPU Policy

If a vGPU policy is not published to tenants, you can delete it from the provider VDC.

Prerequisites

  • Verify that the vGPU policy is not added to an organization VDC. You cannot delete vGPU policies that are available to tenants.

Procedure

  1. From the top navigation bar, select Resources and click Cloud Resources.
  2. In the left panel, select Provider VDCs.
  3. Click a provider VDC from the list.
  4. Under Policies, select the vGPU tab, and select the radio button next to a vGPU policy.
  5. Click Delete.