Configure a vGPU resource pool

Restrictions and guidelines

• The physical graphics card must support the NVIDIA GRID vGPU solution.

• You can use only limited features of the NVIDIA vGPU software without registering a license. To use all NVIDIA vGPU features, deploy an NVIDIA license server.

• A VM can be attached to only one vGPU.

• A physical GPU can be virtualized into only one type of vGPUs. Different physical GPUs on a graphics card can be virtualized into different types of vGPUs.

• A physical GPU cannot work in passthrough mode if it has been virtualized into vGPUs, and vice versa.

• Make sure the graphics card works in graphics mode. If the graphics card works in compute mode, use gpumodeswitch to switch its mode. For more information, see gpumodeswitch User Guide from NVIDIA.

Procedure

Figure-1 vGPU resource pool configuration flow chart

 

Table-1 vGPU resource pool configuration workflow

Procedure	Remarks
Add a vGPU	Required. CAS integrates the NVIDIA Virtual GPU technology to provide GPU virtualization services for VM users. Before creating a vGPU, make sure the host has GPUs that support NVIDIA GRID vGPU installed.
Configure a vGPU resource pool	Required. Perform this task to add vGPUs on different hosts in the same cluster to a vGPU resource pool and add VMs to use the vGPUs to a VM group. The system assigns vGPUs to VMs based on the resource usage in the vGPU resource pool and the VM priority when the VMs in the VM group start or restart.
Install an NVIDIA graphics driver on the VM	Required. Perform this task to install an NVIDIA graphics driver on a VM for the VM to identify vGPUs.