This section provides an overview of the virtual delivery of 3D professional graphics applications and workstations in Citrix Hypervisor. The offerings include GPU Pass-through (for NVIDIA, AMD and Intel GPUs) and hardware-based GPU sharing with NVIDIA vGPU™, AMD MxGPU™, and Intel GVT-g™.
Graphics Virtualization is available for Citrix Hypervisor Premium Edition customers, or customers who have access to Citrix Hypervisor through their Citrix Virtual Apps and Desktops entitlement or Citrix DaaS entitlement. To learn more about Citrix Hypervisor editions, and to find out how to upgrade, visit the Citrix website. For more information, see Licensing.
In a virtualized system, most of the physical system components are shared. These components are represented as multiple virtual instances to multiple clients by the hypervisor. A pass-through GPU is not abstracted at all, but remains one physical device. Each hosted virtual machine (VM) gets its own dedicated GPU, eliminating the software abstraction and the performance penalty that goes with it.
Citrix Hypervisor allows you to assign a physical GPU (in the Citrix Hypervisor server) to a Windows or HVM Linux VM running on the same host. This GPU pass-through feature is intended for graphics power users, such as CAD designers.
Shared GPU allows one physical GPU to be used by multiple VMs concurrently. Because a portion of a physical GPU is used, performance is greater than emulated graphics, and there is no need for one card per VM. This feature enables resource optimization, boosting the performance of the VM. The graphics commands of each virtual machine are passed directly to the GPU, without translation by the hypervisor.
Multiple vGPU enables multiple virtual GPUs to be used concurrently by a single VM. Only certain vGPU profiles can be used and all vGPUs attached to a single VM must be of the same type. These additional vGPUs can be used to perform computational processing. For more information about the number of vGPUs supported for a single VM, see Configuration Limits.
This feature is only available for NVIDIA GPUs. For more information about the physical GPUs that support the multiple vGPU feature, see the NVIDIA documentation.
The following table lists guest support for the GPU, shared GPU, and multiple vGPU features:
|GPU for Windows VMs||GPU for HVM Linux VMs||Shared GPU For Windows VMs||Virtual GPU for Linux VMs||Multiple vGPU For Windows VMs||Multiple vGPU for Linux VMs|
|NVIDIA||YES||YES||YES||YES||YES (see note)||YES (see note)|
Only some of the guest operating systems support multiple vGPU. For more information, see Guest support and constraints.
You might need a vendor subscription or a license depending on the graphics card used.
vGPU live migration
vGPU live migration enables a VM that uses a virtual GPU to perform live migration, storage live migration, or VM suspend. VMs with vGPU live migration capabilities can be migrated to avoid downtime.
vGPU live migration also enables you to perform rolling pool upgrades on pools that host vGPU-enabled VMs. For more information, see Rolling pool upgrades.
To use vGPU live migration or VM suspend, your VM must run on a graphics card that supports this feature. Your VM must also have the supported drivers from the GPU vendor installed.
The size of the GPU state in the NVIDIA driver can cause a downtime of 5 seconds or more during vGPU live migration.
The following restrictions apply when using vGPU live migration:
Live migration of Windows VMs with vGPU enabled from XenServer 7.0 or 7.1 Cumulative Update 2 to Citrix Hypervisor 8.2 is not supported.
Live migration of Linux VMs with vGPU enabled from previous versions of Citrix Hypervisor or XenServer to Citrix Hypervisor 8.2 is not supported.
Live migration is not compatible with GPU Pass-through.
VMs must have the appropriate vGPU drivers installed to be supported with any vGPU live migration features. The in-guest drivers must be installed for all guests using the vGPU feature.
Reboot and shutdown operations on a VM are not supported while a migration is in progress. These operations can cause the migration to fail.
Linux VMs are not supported with any vGPU live migration features.
Live migration by the Workload Balancing appliance is not supported for vGPU-enabled VMs. The Workload Balancing appliance cannot do capacity planning for VMs that have a vGPU attached.
After migrating a VM using vGPU live migration, the guest VNC console might become corrupted. Use ICA, RDP, or another network-based method for accessing VMs after a vGPU live migration has been performed.
VDI migration uses live migration, therefore requires enough vGPU space on the host to make a copy of the vGPU instance on the host. If the physical GPUs are fully used, VDI migration might not be possible.
The following table lists support for vGPU live migration:
|GPRU for Windows VMs||GPU for HVM Linux VMs||Shared GPU for Windows VMs||Virtual GPU for Linux VMs||Multiple GPU For Windows VMs||Multiple GPUs for Linux VMs|
For more information about the graphics cards that support this feature, see the vendor-specific sections of this guide. Customers might need a vendor subscription or a license depending on the graphics card used.
Guest support and constraints
Citrix Hypervisor 8.2 supports the following guest operating systems for virtual GPUs.
Operatings systems marked with an asterisk (*) also support multiple vGPU.
- Windows 10 (64-bit) *
- Windows Server 2016 (64-bit) *
- Windows Server 2019 (64-bit) *
- Windows Server 2022 (64-bit) *
- RHEL 7 *
- RHEL 8 *
- CentOS 7
- CentOS 8
- Ubuntu 16.04 *
- Ubuntu 18.04 *
- Ubuntu 20.04 *
- Windows 10 (64-bit)
- Windows Server 2016 (64-bit)
- Windows Server 2019 (64-bit)
- Windows 10 (64-bit)
- Windows Server 2016 (64-bit)
VMs with a virtual GPU are not supported with Dynamic Memory Control.
Citrix Hypervisor automatically detects and groups identical physical GPUs across hosts in the same pool. If assigned to a group of GPUs, a VM can be started on any host in the pool that has an available GPU in the group.
All graphics solutions (NVIDIA vGPU, Intel GVT-d, Intel GVT-G, AMD MxGPU, and vGPU pass-through) can be used in an environment that uses high availability. However, VMs that use these graphics solutions cannot be protected with high availability. These VMs can be restarted on a best-effort basis while there are hosts with the appropriate free resources.