GRID K2 Windows 2012 r2 non-virtualized server

I’am trying to use GRID K2 on bare metal server HP dl380p gen9. Card was bought with server from hp.
After driver 332.50-quadro-tesla-grid-winserv2008-2008r2-2012-64bit-international-whql.exe installation everything looks fine in device manager but no application is able to use direct3d acceleration.
DxDiag shows dxdiag —> Display —> Device: Microsoft Basic Display Driver
So it’s possible to use K2 on non-virtualized server with direct3d ?

Hi

This may help answer some of your questions:

https://gridforums.nvidia.com/default/topic/403/xenapp-with-nvidia-grid/xenapp-on-bare-metal-with-grid-/

I personally haven’t tried a bare-metal install using a GRID card (I’ve heard of others trying it, but I don’t understand the use case for it, use a Quadro if you want bare-metal), but it should work if configured correctly. Obviously, no vGPU will be available (as there is no virtualisation involved) and how the OS balances between the 2 GPUs (K2 has 2 GPUs), I’m unsure. I’ll have to try it in my lab to see how it performs.

Jason did a very handy video on how to check if it’s configured correctly. Remember, how you connect to it (protocol) will depend on the features made available:

Let us know how you get on.

Regards

Ben

You should also be aware that when using the card(s) without virtualisation you’re relying on the OS not the GRID software to share the load across multiple GPU’s.

In many cases we see that the Windows OS does not do this (particlularly for RDSH) and so the majority of load is placed on a single GPU. This is predominantly an OS limitation and so it is recommended that if you’re planning to use K2 with Windows Server OS you use it in conjunction with a hypervisor.

Thank you all for all the help !
In the end it worked on windows 2012 r2 with drivers downloaded from HP support site.
But now i’am trying to do similar setup but with Windows 2008 r2 sp1. With same drivers and same hardware with no luck 8((
In device manager i can see 2x Nvidia GRID k2 cards, but windows fail to detect displays for each card.
So dxdiag shows only 1 display. On windows 2012 dxdiag/device manager shows 3 displays.
Any suggestions ?

did you enable the RemoteFX in Group Policy?

I am attempting the same thing with Server 2016 and a K1. however I am getting the same 100% load on 1 GPU doesn’t distribute.

Enabling RemoteFX won’t fix it. You’re using the wrong tool for the job.

The Operating System doesn’t have the functionality to use multiple GPUs, and as Nvidia don’t support it for GRID, their driver won’t have the functionality for it either. What you’re seeing is expected, the Operating System will just load up 1 of the GPUs while the others sit there doing nothing.

granted but you will have the GPU and not the Microsoft default driver. you will only be able to use 1 GPU of it.

Hi Jason,

I fully understand that without virtualisation you’re relying on the OS not the GRID software to share the load across multiple GPU’s.
And I agree that Windows OS does not do this and particlularly for RDSH under Windows 2012R2.
But what about Windows 2016 ? it is listed as pre-production beta in https://griddownloads.nvidia.com/flex/GRID_4_Product_Support_Matrix.pdf, is there any pre-production beta driver/grid software ? (with have a Dell R730 with a K1 for testing).

In Virtual GPU Software User Guide :: NVIDIA Virtual GPU Software Documentation there is only documentation about Xen and VMWare. Is there any documentation for RemoteFX vGPU on RDVH/Hyper-V setup or/and Personal Session Desktops configuration (as RDSH is less performant because Windows can’t spread the load across the 4 GPU).
( ref: https://blogs.technet.microsoft.com/hybridcloudbp/2016/11/15/new-rds-capabilities-in-windows-server-2016-for-service-providers/ )

regards