Virtual Graphics Processing Unit (NVIDIA vGPU™) - Virtualization - Wiki - Virtualization - Dell Community

Virtual Graphics Processing Unit (NVIDIA vGPU™)

Virtualization - Wiki

Wikis - Page

Virtual Graphics Processing Unit (NVIDIA vGPU™)

Virtualization - Wiki

Virtual Graphics Processing Unit, or NVIDIA vGPU™, is cutting-edge technology developed by NVIDIA that enables hardware sharing of graphics processing for virtual desktops. This solution provides a sort of hybrid shared mode, where the GPU is virtualized but the virtual machines still run the native NVIDIA video drivers for better performance, and access to a lot more graphics application thanks to OpenGL support. When utilizing vGPU, the graphics commands from virtual machines are passed directly to the GPU without any hypervisor translation. All this is done without sacrificing server performance and so is truly cutting edge.

The Dell PowerEdge R720, which is the recommended server in the Dell DVS reference architecture, can accommodate two graphic cards like the NVIDIA GRID™ K1 or K2 cards. The combination of Dell servers, Citrix vGPU technology and NVIDIA GRID cards enable high-end graphics users to experience high fidelity graphics quality and performance, for their favorite applications at a reasonable cost.

Current Dell DVS (Desktop Virtualization Solutions) offer pass-through (vDGA) and shared (vSGA) graphics solutions with Citrix XenDesktop and VMware ESX Hypervisor. Dell DVS Engineering has tested these technologies on the PowerEdge R720 server platform and the following density numbers were observed with PVS:


Local/ Shared Tier 1   Rack Densities


NVIDIA K1 ( x2)

NVIDIA K2 ( x2)

vSGA   (shared)



vDGA (Pass-Through)




The NVIDIA K1 and K2 cards utilize the NVIDIA Kepler architecture which allows for hardware virtualization of the GPU. So, not only can multiple users share a single GPU, they will get better graphics performance when compared to software virtualization. Multiple K1 and/or K2 cards can be used to support maximum user density on a server.

NVIDIA testing has shown that 16 power users or 32 knowledge workers can be supported on a single K1 card. This translates to 32 power users or 64 knowledge workers supported by the Dell PowerEdge 720 (Dell is still conducting graphics testing to verify this). Launch time frame for this Dell-Citrix-NVIDIA solution is in the later part of the first quarter of 2014. The vGPU solution will be included in the Dell DVS Enterprise Reference Architecture Addendum for Accelerated Graphics – Citrix XenDesktop Reference Architecture in this release.


  2. Dell Reference Architecture
  4. NVIDIA GRID™ main website –
  5. NVIDIA GRID™ K1 & K2 boards -
  6. Citrix Micro-site: