ZStack Cloud 5.3.20>ZStack Cloud Tutorials>ZStack Cloud Private Cloud>vGPU Virtualization Tutorial>Preparations

Preparations

NVIDIA GPU Virtualization

To utilize the NVIDIA GPU virtualization feature, the following preparations are necessary:

Ensure that the pGPU model supports virtualization.
Ensure that the Intel VT-d/AMD IOMMU feature is enabled in the host BIOS, and that the host kernel has IOMMU support enabled.
Ensure that SR-IOV and Memory Mapped I/O above 4GB are enabled in the host BIOS.
Ensure that the pGPU driver, vGPU driver, and VM vGPU driver have been obtained. For more information about driver service or installation methods, please contact GPU device supplier. For recommended driver versions, see Considerations.
For NVIDIA A series users, you have to ensure that your host kernel is upgraded to 4.18 version, and GCC to 8.3.1 version.
To use NVIDIA vGPU software, see NVIDIA Official Documents. Note that you have to purchase the License, set up a License Server, and import the License in advance.
Configure the global settings: Click Settings > Platform Settings > Global Settings. The following global settings are related to the pGPU passthrough feature, you can adjust them as needed:
- PCI Hot-Plugging: Specifies whether to enable hot plugging of GPU devices for a VM instance. The default is true. If a hardware incompatibility error occurs during hot plugging or a hardware device does not support hot plugging, you can set this parameter to false.
- vGPU Default Quota: Specifies the number of GPU devices (including pGPUs and vGPUs) that an account or a project can use. The default is 20.

AMD GPU Virtualization

To utilize the AMD GPU virtualization feature, the following preparations are necessary:

Ensure that the pGPU model supports virtualization.
Ensure that the Intel VT-d/AMD IOMMU feature is enabled in the host BIOS, and that the host kernel has IOMMU support enabled.
Ensure that the host GPU driver and VM vGPU driver have been obtained. For more information about driver service or installation methods, please contact GPU device supplier. For recommended driver versions, see Considerations.
Ensure that the advanced settings VM Instance Hypervisor is set to false, otherwise the VM instance will fail to properly attach the AMD vGPU driver. The advanced setting can be disabled through the following methods:
- For a VM instance: Enter the details page of the VM instance and choose Advanced settings on the top row. Set VM Instance Hypervisor to false for this VM instance.
- For a cluster: Enter the details page of the cluster and choose Advanced settings on the top row. Set VM Instance Hypervisor to false for this cluster.
- For the platform: Click Settings > Platform Settings > Global Settings > Basic Settings. Find VM Instance Hypervisor and set it to false for the entire platform.
Set global settings as needed: Click Settings > Platform Settings > Global Settings. The following global settings are related to the pGPU passthrough feature, you can adjust them as needed:
- PCI Hot-Plugging: Specifies whether to enable hot plugging of GPU devices for a VM instance. The default is true. If a hardware incompatibility error occurs during hot plugging or a hardware device does not support hot plugging, you can set this parameter to false.
- vGPU Default Quota: Specifies the quota of GPU devices (including pGPUs and vGPUs) that an account or a project can use. The default is 20.

Next Chapter: Typical Scenarios

Private Cloud Platform

Private Cloud Platform

ZStack ZSphere Virtualization Platform

ZStack HCI

ZStack Software-Defined Storage

Data Center Management

Edge Orchestration

Cloud-Native Platform

Database Management

Private AI

Advanced Infrastructure Platform

ZStack Cloud Platform

ZStack ZSphere Virtualization Platform

By Scenario

By Industry

Documentation&Tools

Support & Services

Training & Certification

Content

VMware-to-ZStack Case Collection

Preparations

NVIDIA GPU Virtualization

AMD GPU Virtualization