Nvidia K2 pass-through problems on an ESXi 5.1U3 system

[Ashom]

Ok, so I have a new set of Dell R720s with dual Nvidia K2 GPUs on board from Dell.  I've got the latest drivers loaded (340) on the host (ESXi 5.1U3), I can configure the "Enable 3D Support" and I'm able to select "Hardware" under "3D Render" in "Edit Settings".  However, because the software I'm using prefers to access the GPU directly (as much as Windows 7 will let it), I've configured 1 of 4 GPUs as a pass-through device and added it to a VM running Windows 7 w/SP1 and the latest patches.  I've also loaded the latest Nvidia K2 Windows 7 driver (347).  Finally, in order to resolve a conflict between the video drivers, I disabled the VMware vSVGA 3D driver on the Windows side to force it to use the Nvidia K2 driver.

So, here's the problem.  The program still doesn't recognize that the Nvidia GPU is installed on the system (it's based on OpenGL) and I'm unable to start either the "NVIDIA Control Panel" or the "NVIDIA nView Desktop Manager" from the Windows Control Panel.  While I understand that this sounds like a Windows 7 issue (and it may be) I think that at the heart of it all, it's a problem rooted in VMware.

Any help you could provide would be appreciated, thanks!

[Alistar]

Hi there,

the problem is that you have disabled the VMware sVGA 3D Driver. I've been there and done that but because of a different reason - the console had spanned between multiple monitors and had its mouse cursor disappearing and behaving weird. When I tried to use just the nVIDIA monitor that was linked to the GPU, I got no output signal.

You need to re-enable the adapter and reboot the machine. Also please note that the "Enable 3D acceleration" and "PCI Passthrough" are two different things.

• "Enable 3D Acceleration" to enable vSGA (Shared Graphics Adapter) - this will allow you to allocate up to 512MB of Video Memory to a VM and allows for snapshots, vMotions etc. - The 3D will be passed through to hypervisor's xorg daemon who then processes the data and returns the output back to the VM.
• "PCI Passthrough" - Gives you direct access to one of the GRID's GPUs aka vDGA (Dedicated Graphics Adapter) - the hypervisor exposes the whole GPU to the designated VM and - everything is processed by whole GPU core and passed through directly to the VM.

The thing is however, that the VMware "monitor" is used as the "base" for your VM's graphics - these two options above are "enhancing" it, so you will need to keep it in. I'd suggest you to look at VMware Horizon VDI with its PCoIP protocol finely tuned for just this kind of applications.

Cheers!

[Ashom]

So the short version is that I have to purchase Horizon with View to be able to dedicate a GPU to the VM.  Otherwise I'm limited to the VMware vSGA driver.

[Alistar]

Pretty much this or Citrix. I have tried Horizon/View today in a lab environment and it works flawlessly - however navigating through virtual console is a huge pain because of the mouse cursor disalignment The VMware adapter has to be present unfortunately. So more or less - yeah to fully draw on vDGA you have to have a software that can translate multiple monitors for a remote session correctly.

[admin]

R u still hitting the issue ?

Do you have logs available now ?