I try to use passthrough functionality with a Tesla M10 but I meet some issue.
I have a DL380 G9 with a Tesla M10. I have installed VMware ESXi 6.0.0U2 on it and I deployed a Windows 7 pro virtual machine.
With VSphere client, on "Advanced Parameter" I activate passthrough. I assign on GPU on my VM and I restart it.
Unfortunately, the passthrough does not work.
Indeed, I have a monitor plug to the DL380, but I only see ESXi interface, and not the desktop of Windows 7 VM.
I think that a NVidia license is needed to use this functionality, or to activate it, but I don’t know where use this license.
So, I have some news about my issue.
I found the good .vib file to install on my ESXi.
After this installation Tesla M10 card is well detected on VSphere.
Now, I try to install driver on Windows 7 VM but no compatible material is detected…
first of all I would recommend to have a closer look into our deployment guides. What are you trying to achieve?
Passthrough doesn’t require the vGPU manager! And if you want to deploy vGPU you need the correct vGPU manager/driver combination from the GRID portal. Please be aware that this requires a software license and is not available on the public NV download page.
If you could describe what you want to achieve I’m happy to guide you what to do and where to get the required resources.
The aim of the manipulation is to redirect Windows 7 VM display to the KVM, plug on DL 380 VGA connection, thanks to passthrough functionality.
I already installed .vib on my server, I assign new PCIe component to my VM (with VSphere client) but when I try to install NVidia driver, I have error message which tell that "no compatible material detected).
which driver did you try? You need to use a Tesla capable driver for M10.
Create an account here: Retrieve token from server append to link and download the driver package for your hypervisor. As I said you require to buy software licenses for your use case.
I installed NVIDIA-vGPU-VMware_ESXi_6.0_Host_Driver_367.106-1OEM.600.0.0.2494585 on my ESXi
On my VM with Windows 7, I installed 370.12_grid_win8_win7_server2012R2_server2008R2_64bit_international
After this, I obtained a black screen on my VM console. I de-activate passthrough, relaunch the VM, de-activate VMWARE SVGA driver, shutdown Windows and re-put the passthrough.
Now, my console work well.
But, to see Windows, I use VSphere client.
A monitor is plug on the HP proliant and I would like to see Windows on this monitor.
Concerning the licence, we bought them when we buy the HP proliant so I enter them on the previous link :).
Well this is expected behavior. As soon as the vGPU profile is assigned you won’t be able to administer the VM with the VMWare console. You need to use RDP or any other remoting protocol like Citrix/VMWare.
Yes I understand how virtualization works.
But for one of very specific use, we need to see the VM through the graphic card.
On another computer, with quadro K2000 GPU, I configure passthrough on the VM (with success) and plug my monitor on this card. And I see the VM desktop.
But with tesla card, as there is no way to plug monitor on it, I suppose that is not possible.
This is correct, Tesla GPUs do not have the ability to accept a physical monitor connection. The GPU you are referring to (the K2000) is a Quadro GPU. It is not supported in ANY servers, as it is designed to work in a PC / Workstation.
If you would like to see the output of the M10, you’ll need to use a remoting protocol of your choice. Slightly off topic, but if you were using XenServer, you’d be able to use the Hypervisor console, but ESXi does not support this functionality.
I have been using passthrough with Nvidia GRID K1 to basically link the 4 GPUs to 4 VMs. Then connect to the VMs using VMware Horizon(Blast protocol). It was working fine and I get 3D rendering etc… I have never needed any GRID licencing to do this. Well, I have opted to use vDGA since vGPU was not an option for Linux VMs with GRID K1, then I kind of liked using vDGA since I think it provides better performance.
I am planning a hardware refresh, and according to a quick search, the best alternative for the GRID K1 is the Tesla M10 (4GPU and 32Gb memory). Is it possible to replicate the same setup using M10 without the need for a licence? Basically, add the GPU PCI device to the VM, Install the generic Nvidia driver and render 3D objects.