Support statement regarding NVIDIA vGPU + XenApp 7.5

am trying to figure out if there is a document somewhere that state that the combination NVIDIA vGPU + VDA 7.X for server (XenApp) is officially supported by Citrix / NVIDIA.

I know it works but I am running into some trouble demonstrating internally this is officially supported.

For some reason some of our folks are under the impressions that
NVIDIA vGPU is only supported under Win 7.x + VDA for desktop
VDA 7.x server only supports pass-through.


Hi Didier,
No. XenApp is not comnpatible today with vGPU. Only XenServer + XenDT for vGPU.
You can pass thru the entire GPU to a WinServer VM and use XenApp that way. One GPU per WinServer OS VM tho.

Thanks Steve. Appreciated. We will reconfigure our servers and have only two XenApp VMs per R720 + Grid K2 for the time being.


I may have to correct you.

NVIDIA have released driver 332.83 which is released for Operating System:
Windows Server 2008R2
Windows Server 2012
Windows Server 2012R2

Citrix XenApp 7.x is build for Windows Server 2008R2/2012/2012R2.

I have tried all the vGPU profiles on GRID K1 & K2 and it works.

Can you please confirm that NVIDIA/Citrix supports XenApp 7.x with Server 2008R2/2012/2012R2 with vGPU ?

Thomas Poppelgaard

XenApp is more than just a Server 2008R2, vGPU is supported and QA’d for VDI only at the moment. I expect it could well work by default but it’s not a configuration Citrix or NVIDIA test. Really because although people have asked for this when queried we’ve never found a user case. XenApp offers a multi-user sharing mechanism already - why would you attach only part of a vGPU to server VM to server multiple users, rather than increase the density of users… Multiple GPUs to a server VM is more interesting scenario imo… if someone can come up with a XenApp vGPU scenario it’s something we’d consider but I’ve yet to understand one as yet.


Rachel, having seen Thomas’s demo I am curious about education use cases such as K-12 (US reference) schools? Lots of students using simple graphics based apps, even just Google Earth? Really any bulk quantity of simple GPU use users, take a large call center using a web delivered LoB app? Based on Thomas it looks like you could get very high densities?

I wonder if putting too many users on a single engine GPU passthrough is really more efficient than assigning individual vGPUs in the case of a XenApp server. From my tests, it seems that the VM itself would probably benefit more, and doubling up seems to not help in a lower-demand case, at least for XenDesktop. Pure XenApp is a bit of a different scenario.

It should be easy enough to test given enough real or simulated connections. I agree with Luke that educational institutions make use of many more simple programs, like Google Earth, ArcGIS and the like that can benefit from accelerated graphics, but do not have the high demands of CAD-CAM applications.

Lets use an Office 2013 workload as an example. It performs considerably better (from an end user standpoint and server resource standpoint) with a GPU due to its reliance on DirectX10 in our testing environment. The simple fact is that in terms of resources it seems to make a lot more sense to use a vGPU instead of passthrough because we are not limited to 4 VMs per K1 card.

To follow up with Luke in regards to educational institutions we have hundreds of smaller applications that would benefit from DirectX10+OpenGL 4 in our XA environment. These applications are spread across a number of traditional worker groups. Passthrough is not an effective use of server resources. I cannot understand why Citrix (I am politely pointing my virtual finger at Rachel) doesn’t think that there is a use case here. Why would I use XD for light loads where XA is the better choice?

Kevin Miller

XA will most definitely scale better and be cheaper in many ways to run with some additional flexibility. Two XenApp servers running GPU passthrough with a K2 engine mapped to each should provide a lot more flexibility than mapping individual vGPUs to a more limited number of potential users. I’m completely with your line of thinking, Kevin.

Hey Thomas and Dieder,
vGPU may work just fine in a XenApp setup no question as Thomas has indicated. However, it is not QA’d or supported at this time. We are not promoting it.
Hope that clarifies things a bit.

Sorry, my initial misstatement, so there no XenApp involved here, but this is perhaps still of interest.

We are doing some testing with a GRID K1 in passthrough mode with a dental hygiene application that requires DirectX and is being served via a VM configured as a Win 2008 R2 server to multiple users (sorry, the package isn’t supported yet on Win 2012 R2) and it seems to be working really well. Have not worked out the scaling yet, but one Windows server run with 8 VCPUs and 16 GB of memory as a VM on a XenServer on an R720 will hopefully be able to handle 12 or more sessions compared to what would take each XenDesktop VM a vGPU and 4-6 GB of memory. Three concurrent instances barely use together about 1 GB of extra RAM and there is still a lot of CPU power available (it hit around 5%). I’m guessing we should be able to scale this to handle otherwise what would be close to 20 sessions (which would otherwise take 80 - 120 GB of RAM and 40 VGPUs with XenDesktop VMs). We are, BTW, running a XenApp 7.5 instance on the same physical server, also with a GPU passthrough configured, so making good use of the box with a variety of resources.

Just an update as Citrix have recently changed their stance on supporting XenApp with vGPU

Hi Tobias

just two questions about xenapp 7.6 and applications running inside that terminalserver 2008 on a xenserver 6.5 with a K2 Nvidia vgpu setup.

citrix support until now, does not know what to do?


  • in a rdp or ica session, nvidia-smi.exe -l output shows me "after changing some settings" use of gpu from chrome and firefox.
  • same server login via xencenter in console session output smi.exe shows me Office using gpu. in a rdp or ica session office runs on cpu. is there something i missed?


  • we publish full desktop byod to our employes. and from inside that desktop - users connecting to different costumers via rdp or ica. Does the rdp and/or ica protocol can takes benefits from the Nvdidia vgpu?

thank you Tobias in advance