NVIDIA
Information about vGPU technology and virtualization deployment
Hi vGPU forum, I'm investigating how to deploy a configuration to virtualize into an unique server, several virtual machines all running at the same time, and all compliant with OpenGL >= 3.2 version, according to our SW 3D graphic engine requirement. Collecting a few information, it seems that I could deploy this solution using vGPU technology (NOT using PASS-THROUGH technology because only 1 machine should access to the video card, while I want that ALL virtual machines shall access to video card resources). Reading from nVIDIA GRID user guide http://us.download.nvidia.com/Windows/Quadro_Certified/GRID/370.21/ESXi-6.0/367.124-370.21-grid-vgpu-user-guide.pdf, it seems that we could use Tesla M10 or Tesla M60 to obtain 3D graphic performances for our SW (engineering CAD application - like), in details we detect M10-4Q or M10-2Q as Tesla M10 vGPU types, and M60-2Q or M60-1Q as Tesla M60 vGPU types. Virtualization systems we're investigating are: [list] [.] Citrix XenServer [/.] [.] VMware Hypervisor ESXi [/.] [.] Windows Server 2016 with Hyper-V role [/.] [/list] My questions are: 1) I haven't well understood if it's possible to deploy vGPU for Windows Server 2016 (NOT vDGA - PASSTHROUGH technology, I need vGPU to generate several vGPU to be assigned to each VM starting from 1 physical GPU). If so, what are the requirements? 2) reading this User Guide http://us.download.nvidia.com/Windows/Quadro_Certified/GRID/370.21/ESXi-6.0/367.124-370.21-grid-vgpu-user-guide.pdf I don't see the vGPU technology applied to Windows Server, only PASSTHROUGH, but here https://docs.microsoft.com/en-us/windows-server/remote/remote-desktop-services/rds-graphics-virtualization it seems that vGPU on Windows Server is feasible using RemoteFX vGPU. Where is the truth? Is there a clear user guide regarding it? 3) Here https://social.technet.microsoft.com/Forums/SECURITY/en-US/0b87ddc9-efdc-4447-90b3-ab2b725e88f0/server-2016-remotefx-tesla-m60-support?forum=winserverTS it's written that "If you are planning to use CAD applications, you may want to consider moving to a hypervisor that supports vSphere or XenServer which will support the GRID Virtual Workstation licenses. The vDWS licenses contain our Quadro tools which are used to enhance CAD programs, especially when rendering large or 3D designs." So this means that Windows Server 2016 with Hyper-V role, cannot support GRID vDWS license (using Tesla M10 or Tesla M60)? 4) Generally speaking, using nVIDIA GRID layer with vGPU technology, is it possible to use Windows Remote Desktop Connection to connect from another client to the VM without losing vGPU 3D performances? Or a VDI client tool is required? If so, VMware offers Horizon, Citrix offers XenDesktop , what offers Microsoft? Any replies and comments will be really appreciated. Best Regards, Francesco
Hi vGPU forum, I'm investigating how to deploy a configuration to virtualize into an unique server, several virtual machines all running at the same time, and all compliant with OpenGL >= 3.2 version, according to our SW 3D graphic engine requirement.

Collecting a few information, it seems that I could deploy this solution using vGPU technology (NOT using PASS-THROUGH technology because only 1 machine should access to the video card, while I want that ALL virtual machines shall access to video card resources).

Reading from nVIDIA GRID user guide http://us.download.nvidia.com/Windows/Quadro_Certified/GRID/370.21/ESXi-6.0/367.124-370.21-grid-vgpu-user-guide.pdf, it seems that we could use Tesla M10 or Tesla M60 to obtain 3D graphic performances for our SW (engineering CAD application - like), in details we detect M10-4Q or M10-2Q as Tesla M10 vGPU types, and M60-2Q or M60-1Q as Tesla M60 vGPU types.

Virtualization systems we're investigating are:

  • Citrix XenServer

  • VMware Hypervisor ESXi

  • Windows Server 2016 with Hyper-V role


My questions are:

1) I haven't well understood if it's possible to deploy vGPU for Windows Server 2016 (NOT vDGA - PASSTHROUGH technology, I need vGPU to generate several vGPU to be assigned to each VM starting from 1 physical GPU). If so, what are the requirements?

2) reading this User Guide http://us.download.nvidia.com/Windows/Quadro_Certified/GRID/370.21/ESXi-6.0/367.124-370.21-grid-vgpu-user-guide.pdf I don't see the vGPU technology applied to Windows Server, only PASSTHROUGH, but here https://docs.microsoft.com/en-us/windows-server/remote/remote-desktop-services/rds-graphics-virtualization it seems that vGPU on Windows Server is feasible using RemoteFX vGPU. Where is the truth? Is there a clear user guide regarding it?

3) Here https://social.technet.microsoft.com/Forums/SECURITY/en-US/0b87ddc9-efdc-4447-90b3-ab2b725e88f0/server-2016-remotefx-tesla-m60-support?forum=winserverTS it's written that "If you are planning to use CAD applications, you may want to consider moving to a hypervisor that supports vSphere or XenServer which will support the GRID Virtual Workstation licenses. The vDWS licenses contain our Quadro tools which are used to enhance CAD programs, especially when rendering large or 3D designs." So this means that Windows Server 2016 with Hyper-V role, cannot support GRID vDWS license (using Tesla M10 or Tesla M60)?

4) Generally speaking, using nVIDIA GRID layer with vGPU technology, is it possible to use Windows Remote Desktop Connection to connect from another client to the VM without losing vGPU 3D performances?
Or a VDI client tool is required? If so, VMware offers Horizon, Citrix offers XenDesktop , what offers Microsoft?

Any replies and comments will be really appreciated.

Best Regards,

Francesco

#1
Posted 04/13/2018 08:58 AM   
Hi Francesco, these are a lot of questions. First of all I would recommend to use our docs.nvidia.com/grid site to get the proper documentation and all information you need. Anyway I try to respond to your questions: 1) For sure we also support Server OS with vGPU. This is our vApps licensing but vApps doesn't contain Quadro features and therefore this is not recommended for 3D applications. 2) Hyper-V doesn't support vGPU technology yet. RemoteFX is something different, a such called API intercepting technology that has restrictions and is not really feasible for 3D applications. 3) Correct, currently you should consider XenServer or vSphere to use vGPU. Hyper-V only supports such called DDA = Passthrough. If you would DDA on Hyper-V you can use QvDWS licensing but your VM density is restricted to the GPUs available for example you could use 2 VMs with 1 M60. 4) You can use RDP but you may see less performance compared to Citrix or VMWare. Especially in low bandwidth environments there is still a huge difference between the protocols. Hope this helps for your further investigation. Best regards Simon
Hi Francesco,

these are a lot of questions. First of all I would recommend to use our docs.nvidia.com/grid site to get the proper documentation and all information you need. Anyway I try to respond to your questions:

1) For sure we also support Server OS with vGPU. This is our vApps licensing but vApps doesn't contain Quadro features and therefore this is not recommended for 3D applications.

2) Hyper-V doesn't support vGPU technology yet. RemoteFX is something different, a such called API intercepting technology that has restrictions and is not really feasible for 3D applications.

3) Correct, currently you should consider XenServer or vSphere to use vGPU. Hyper-V only supports such called DDA = Passthrough. If you would DDA on Hyper-V you can use QvDWS licensing but your VM density is restricted to the GPUs available for example you could use 2 VMs with 1 M60.

4) You can use RDP but you may see less performance compared to Citrix or VMWare. Especially in low bandwidth environments there is still a huge difference between the protocols.

Hope this helps for your further investigation.

Best regards

Simon

#2
Posted 04/16/2018 06:10 AM   
Scroll To Top

Add Reply