VDI Machines with NVIDIA Tesla T4 Profile are stuck when using Horizon Client

Hello everyone…
I’m in the middle of a POC with Horizon VDI.
We are using the New Tesla T4 GPU card.
When i connect to a machine with the VMware Agent I cannot actually work on the machine, it freeze after 1 minute. On the HTML version it’s not happening.
On the other hand - i noticed that when i’m using the client to connect to a machine without GPU profile, it’s working, so it seems something with the driver maybe.

I’m running the latest VMware Horizon 7.8.0
vCenter and ESXi 6.7 U1
8GB RAM for each VM (The option of “Reserve all memory - all locked” is enabled by best practice)

What it could be?
I’ll appreciate your help.

Thanks.

Which OS/version? Which vGPU version?

The VM machines OS is Windows 10 1803.
Driver installed on the VM’s OS: 412.31_grid_win10_server2016_64bit_international
VIB installed on the ESXi: NVIDIA-VMware_ESXi_6.7_Host_Driver-410.107-1OEM.670.0.0.8169922.x86_64

An update:
It’s happening while i’m using the Blast protocol only. when I’m using PCoIP protocol it works fine.
I will add the Horizon GPO ADMX to my DC and i’ll try to debug the blast.

Hey @MaorZ,

I’m having the same problem with Windows 10 2016 LTSC, Horizon 7.8, and the 412.31 vGPU driver (m10-1q profile)… Did you find a resolution?

Thanks,
Justin

Hi Guys,

We running a POC with a Tesla T4 card.
Running on vmware 6.7 Update 2 and using at the moment the latest windows 2016 and vsphere 6.7 driver from nvdia
(NVIDIA-GRID_vSphere-6.7-418.66-418.70-425.31)

When i start an VM (windows 2016) and a gpu profile the esx host will crash on a daily base (PSOD), panic requested by another PCPU and many 0x45 nr nvidia showed up.

When i start the VM without the GPU profile attached on the Vm it will keep running.

A known issue for the VM thats start with an installed nvidia driver in the OS , is that the console session turn black.

somebody else having those issues or does somebody have a solution?

thanks

@brammetje I’d recommend you start a new thread as what you are describing sounds unrelated to this thread.

In regards to this thread… I’ve been working with Nvidia and VMware support for many weeks now on this issue. I could replicate the screen freeze on initial Blast Extreme connection and subsequent reconnects with the 7.1 (412.16) and 7.2 (412.31) versions of the vGPU driver and Horizon 7.8 Agent (on both Windows 10 2016 LTSC and 2019 LTSC).

Nvidia has provided a workaround that appears to work.
Create the following registry setting on the master image:
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\services\nvlddmkm]
"NVFBCEnable"=dword:00000001

Based on Nvidia’s very detailed response to me (which I appreciate), it sounds like without this registry setting, the vGPU driver on first Horizon session thinks it’s not installed properly and does a reload which causes the screen to freeze… At which time a reboot of Windows is required to finish the driver reload… But in the case of my non-persistent instant clones, the VM is deleted when the user logs out so every connection is a first session. In my case, once I created the above registry setting on my master image, I no longer experienced the screen freeze on initial connection or the black screen on subsequent reconnects.

Hopefully this is helpful to someone else.
Thanks,
Justin

I can confirm this is still and issue, but the registry entry vdiguywi suggested works in the following environment:

Windows 10 1903 64-bit
Horizon Agent 7.9
NVIDIA driver version 431.02

Worked for me, M10, Windows 10 1903, Horizon 9. Thanks for sharing @vdiguywi

Anyone ever get to the bottom of why this occurs? We are experiencing this issue but only on our T4s/P4s and not on our M10s. Any configuration changes we need to make?

Hi Zdawson

This is quite an old thread, can you confirm which vGPU, Horizon and OS versions you’re running?

vGPU 11.0 was released last week, if you’re not already running it, I’d suggest you upgrade to that as a start and see if it helps.

Regards

MG