vsphere licensing NVIDIA vGPU

[Bertrand8919]

Hello everyone,

We don't have experience with vmware professional solutions but we intend to implement a virtualization platform on a HP/Dell/Supermicro server having a NVIDIA Tesla P40 GPU card. We thought to base this solution on vSphere.

A typical user’s case would be as follows:

• 1 x VM with Windows Server is running a 2D/3D application accessed by a remote user through Remote Desktop. For this user, I need to allocate a part of the GPU resources.
• 1 x VM with a Linux Distribution (Scientific Linux/CentOS/Ubuntu)  is accessed by a remote user through SSH in order to run a simulation/deep learning algorithm that will use the remaining  GPU ressources.

To make this scenario real, it seems I could use:

1) The vGPU software with one of vgpu license (Grid vPC or Quadro vDWS) with the correct vSphere license (Enterprise ?)

2) the vSGA from vSphere, but it seems that I need to buy (?) a specific driver from NVIDIA for that

My questions are as follows :

1) Which vSphere license do I need to support the vgpu : Standard or Enterprise?

2) Is vSphere standard enough to share GPU ressources with vSGA?

3) What is the best solution (between vSGA and vGPU) for the user's case described above?

Thank you in advance for your help!

[batuhandemirdal]

Hello,

1- You will need an Enterprise vSphere license.

2- GPU virtualization came in, allowing the VM to simultaneously share a single physical GPU. This feature, called vSGA (Virtual Shared Graphics Acceleration), is compatible with all major VMware technologies, including vMotion, and runs virtual GPU-powered virtual desktops. Hardware-accelerated 3D graphics support delivers significantly higher performance in a 3D environment.

3- With vDGA and vSGA it is partly possible to manage the VDI desktop repository but you also need to purchase VMware Horizon View.

One of the biggest changes is the new licensing model. There is now a need for an NVIDIA license server for vGPU use. After the NVIDIA license server is installed on the Windows server in the video, this server has been registered on the NVIDIA license portal and the license number assigned to it has been assigned and the license file downloaded has been installed on the windows license server installed. The NVIDIA driver settings of the Virtual Desktops show the license server. Otherwise use is limited to 3fps.

All information and software related to the installation is accessed through the license portal (at least I haven't found any other methods yet)

Another important point is that Tesla cards come in a mode for the HPC that we call Compute Mode. For horizon / vGPU installations, it is necessary to convert the vSphere boot with an iso file provided by the NVIDIA installation files and the utility with the utility into the Graphic mode. Graphic mode provides the following information:

GPU passthrough with hypervisors. At the time of publication, this includes Citrix XenServer 6.2, 6.5, VMware ESXi 5.1, 5.5, 6.0, Red Hat Enterprise Linux 7.0, 7.1.

GPU passthrough to Windows VMs on Xen and KVM hypervisors

GRID Virtual GPU deployments.

VMware vSGA deployments.

good work

https://www.batuhandemirdal.com.tr/

[Bertrand8919]

Thank you for the detailed reply!

Let's check if I understood correctly.
If I want to use the vGPU software for sharing resources of my Tesla card between VMs (no passthrough) I need only vSphere Enterprise  + NVIDIA vgpu license (Grid vPC or Quadro vDWS) and nothing else. I do not need vmWare Horizon.

If I want to use vSGA or vDA I need vSphere Standard + vmWare Horizon.

Best,

[batuhandemirdal]

Yes, it is true.

I recommend that you review the links below.

VMware Compatibility Guide - vsga

https://www.vmware.com/content/dam/digitalmarketing/vmware/en/pdf/partners/nvidia/vmware-nvidia-grid...