You can allow an OpenStack instance to use an NVIDIA GRID vGPU device on its ESXi host.

For more information about cleaning the stale resource provider and vGPU trait, see Clean Stale vGPU Resource.
Note:
  • Only one vGPU is supported per OpenStack instance.
  • The same vGPU profile is used for all OpenStack instances.

Prerequisites

Verify that the driver for your vGPU device is installed on the ESXi host.

Procedure

  1. Log in to the Integrated OpenStack Manager as the root user. 
    ssh root@mgmt-server-ip
  2. Edit the Nova compute configuration. 
    viocli update nova-compute
  3. In the vmware section, add the gpu_profile parameter and set its value to the vGPU profile that you want to use.
  4. Add the profile_fb_size_kb parameter and set its value to the size of the vGPU frame buffer in kilobytes (KB).

    For example, enter profile_fb_size_kb: 4096 to indicate a frame buffer of 4096 KB.

    conf:
  nova_compute:
    vmware:
      gpu_profile: grid_t4-2q
      profile_fb_size_kb: 4096

    For more information about frame buffers, see the NVIDIA Virtual GPU Software User Guide.

  5. Log in to the VMware Integrated OpenStack dashboard as a cloud administrator.
  6. Select the admin project from the drop-down menu in the title bar.
  7. Select Admin > Compute > Flavors.
  8. Create a new flavor or choose an existing flavor to use for NVIDIA GRID vGPU passthrough.
  9. Select Update Metadata next to the flavor that you want to use.
  10. In the Available Metadata pane, expand VMware Driver Options for Flavors and click the Add (plus sign) icon next to VMware vGPU.
  11. Set the value of vmware:vgpu to 1 and click Save.

What to do next

You can now configure an instance to use an NVIDIA GRID vGPU by configuring the instance with the flavor that you modified in this procedure.