Nvidia M10-1B GRID driver issue - Installing in Platform Layer - Problems with CTX241448

[Rob Jaudon1709158946]

All,

 

When attempting to install NVidia GRID M10-1B drivers on a brand new Platform layer we are following the CTX article https://support.citrix.com/article/CTX241448

by adding the below registry keys.

 

Registry keys as per CTX241448

Add a value called AlwaysOnBoot (if it doesn't already exist) as a multi-string value.  Add the following two lines:

C:\Program Files\NVIDIA Corporation\

C:\Program Files (x86)\NVIDIA Corporation\

 

After adding the above keys and rebooting the platform layer machine the directories disappear and the GRID driver control panel is non-functional.  I have found if I remove the string values in the registry the folders reappear post reboot and the GRID driver control panel works just fine.

 

For example,

Add above registry entries, then reboot, some of the subfolders disappear.  Control panel EXE is not there in dir - C:\Program Files\NVIDIA Corporation\Control Panel Client\nvcplui.exe

 

If I delete the registry entries and reboot the folder structure reappears and full functionality is there and the control panel EXE is present at C:\Program Files\NVIDIA Corporation\Control Panel Client\nvcplui.exe.

 

Normal install where everything works with no registry entries in place

 

 

With Registry entries added to [HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\unifltr]

 

 

NOTE: If I delete the registry entries and reboot, the folder structure reappears and full functionality is there and the control panel works

 

 

What is the proper installation method for the AlwaysOnBoot registry entries?

Should you add the registry entries prior to installing the GRID drivers? Or Post Driver install by prior to reboot to complete the driver install?

 

My workflow has been this...

 

Get new platform layer

Install GRID driver

Add AlwaysOnBoot registry entries

Reboot

Issue occurs

 

I've tried these different install options

 

Get new platform layer

Install GRID driver

Add AlwaysOnBoot registry entries

Reboot

Issue occurs

 

*************************************

Get new platform layer

Add AlwaysOnBoot registry entries

Install GRID driver

Reboot

Issue occurs

 

*************************************

Get new platform layer

Add AlwaysOnBoot registry entries

Install GRID driver

Reboot

Issue occurs

Check AlwaysOnBoot registry entries

Reinstall Nvidia GRID driver

Seems to work

 

*************************************

 

Get new platform layer

Add AlwaysOnBoot registry entries

Reboot

Install GRID driver

Reboot

Issue occurs

At this point clearing the registry entries and rebooting does not bring functionality like in past tests

  

 

Environment Notes

Vmware Vsphere 6.5                

Nvidia Driver  - NVIDIA-GRID-vSphere-6.5-390.113-390.115-392.37

ELM Version 19.02.017

Citrix 7.15 LTSR CU2

Win10 1607

 

Any help or suggestion would be greatly appreciated.

 

Rob Jaudon

[Richard Buffone]

You mentioned if you remove the AlwaysOnBoot keys then the control panel works. Are you seeing any Nvidia control panel issues on your published image at all if you don't follow the CTX article?

[Rob Jaudon1709158946]

58 minutes ago, Richard Buffone said:

You mentioned if you remove the AlwaysOnBoot keys then the control panel works. Are you seeing any Nvidia control panel issues on your published image at all if you don't follow the CTX article?

 

richardbuf,

 

I haven't got that far yet.  This is just the Platform Layer, GRID install with the registry entries (CTX241448 article), and a reboot. 

 

In the past prior to the CTX241448 article many months ago, yes I was seeing issues with the NVidia Control Panel on the published image.

 

Rob

[Richard Buffone]

I would suggest testing out the layer without adding the AlwaysOnBoot keys and see if any problems occur on the provisioned machine.

 

Did you set this key on the OS layer from the prior time you set this up? I find it odd the issue returns if leave the keys in place. The AlwaysOnBoot key basically tells our filter driver to not virtualize those folders, in which case it should work as if our filter wasn't even installed.

[Rob Jaudon1709158946]

Richardbuf, I originally did publish a full image without the registry keys and the issue with the Nvidia Control Panel was still present.

It was the same state I have been in for at least 6 months.  I originally worked with Gunther on this issue at the end of last year.

 

 

[Thomas Rodgers]

Had a similar problem with our deployment, I believe this was the order we did to achieve success in the final layer

 

Get new platform layer

Install GRID driver ( if you need to setup GRID licence server, add it via Registry key if the control panel won't open) https://docs.nvidia.com/grid/latest/grid-licensing-user-guide/index.html#windows-registry-grid-license-settings

Add AlwaysOnBoot registry entries

Reboot

Files will be missing and control panel won't work, it's fine as once the layer is published this will reappear

Finalise layer and publish into an image

Once image is published test and make sure the driver is working and the control panel opens

[Rob Jaudon1709158946]

 

 

1

13 hours ago, Thomas Rodgers said:

Had a similar problem with our deployment, I believe this was the order we did to achieve success in the final layer

 

Get new platform layer

Install GRID driver ( if you need to setup GRID license server, add it via Registry key if the control panel won't open) https://docs.nvidia.com/grid/latest/grid-licensing-user-guide/index.html#windows-registry-grid-license-settings

Add AlwaysOnBoot registry entries

Reboot

Files will be missing and control panel won't work, it's fine as once the layer is published this will reappear

Finalize layer and publish into an image

Once image is published test and make sure the driver is working and the control panel opens

trodgers1,

 

Thank you for this.  I am definitely seeing that behavior but have not performed an image push because at this stage I thought it was broke.  

I will definitely test this out and update you.

 

Thank you for the update.

[Mike Kelly1709153237]

What version of the Nvidia software are you running?  We had a similiar issue but the reverse.  With Nvidia GRID software 6.2, we did NOT need the AlwaysOnBoot reg key.  However, when we upgraded to the latest 8.0... we could not get the Control Panel to function without it.  Also, fun fact... the registry key has to be laid down and then a reboot applied BEFORE you install the nvidia grid drivers otherwise the Nvidia control panel will fail to open.

[Rob Jaudon1709158946]

On 6/14/2019 at 3:39 PM, Mike Kelly1709153237 said:

What version of the Nvidia software are you running?  We had a similiar issue but the reverse.  With Nvidia GRID software 6.2, we did NOT need the AlwaysOnBoot reg key.  However, when we upgraded to the latest 8.0... we could not get the Control Panel to function without it.  Also, fun fact... the registry key has to be laid down and then a reboot applied BEFORE you install the nvidia grid drivers otherwise the Nvidia control panel will fail to open.

 

 

Mkelly439,

 

We are running version Nvidia Grid Graphics Driver 392.37 which is tied to 6.4.

I have tried both with and without the AlwaysOnBoot key and either way I am just not getting a working product post image push.

[Mike Kelly1709153237]

4 minutes ago, Rob Jaudon1709158946 said:

 

 

Mkelly439,

 

We are running version Nvidia Grid Graphics Driver 392.37 which is tied to 6.4.

I have tried both with and without the AlwaysOnBoot key and either way I am just not getting a working product post image push.

We just followed this article to configure our platform layer with Nvidia.

 

https://support.citrix.com/article/CTX225997

 

 

I can't speak for v6.4 but I know 6.2 and 8.0 works as that's what we've tested!

 

[Rob Jaudon1709158946]

Just a quick update.

I have followed the Mkelly439 posted https://support.citrix.com/article/CTX225997 and done the following:

 

NOTE:  I did not use the AlwaysOnBoot registry entry in this test as in article CTX241448.

1. Created a new PL 
2. Added the Shared PCI device on the PL VM
3. Installed the Nvidia GRID driver - NVIDIA-GRID-vSphere-6.5-390.113-390.115-392.37
4. Installed VDA - Citrix_7_15_CU2
5. Installed 2 VDA Hotfixes for CU2
6.     DesktopVDACoreWX64_7_15_2001
7.     profilemgt_x64_7.15.2001
8. Joined PL machine to the domain
9. Logged on as network user, rebooted twice, logged on as admin and deleted network user profile
10. Installed Citrix CQI - 1.4
11. Rebooted
12. Installed PVS tools - Citrix_7_15_CU1
13. Rebooted
14. Finalized
15. Published image.

At each step of the process, I verified the GRID drivers were functional.

 

After publishing the image on the first boot the image came up and was "Getting Devices Ready" and proceeded to go into a boot loop.

As a test...Shutdown the image and in PVS changed it to PRIVATE and started VM to see if it corrected itself.

It got past the Getting Device Ready page and came up in the console to where I could see the logon.  At this point, I should not be able to see the console because the GRID driver takes over.  I would need to RDP into the VM to check on things.

RDP'd into the device and it was non-functional.  Could not launch the Nvidia Control panel.

[Mike Kelly1709153237]

21 hours ago, Rob Jaudon1709158946 said:

Just a quick update.

I have followed the Mkelly439 posted https://support.citrix.com/article/CTX225997 and done the following:

 

NOTE:  I did not use the AlwaysOnBoot registry entry in this test as in article CTX241448.

1. Created a new PL 
2. Added the Shared PCI device on the PL VM
3. Installed the Nvidia GRID driver - NVIDIA-GRID-vSphere-6.5-390.113-390.115-392.37
4. Installed VDA - Citrix_7_15_CU2
5. Installed 2 VDA Hotfixes for CU2
6.     DesktopVDACoreWX64_7_15_2001
7.     profilemgt_x64_7.15.2001
8. Joined PL machine to the domain
9. Logged on as network user, rebooted twice, logged on as admin and deleted network user profile
10. Installed Citrix CQI - 1.4
11. Rebooted
12. Installed PVS tools - Citrix_7_15_CU1
13. Rebooted
14. Finalized
15. Published image.

At each step of the process, I verified the GRID drivers were functional.

 

After publishing the image on the first boot the image came up and was "Getting Devices Ready" and proceeded to go into a boot loop.

As a test...Shutdown the image and in PVS changed it to PRIVATE and started VM to see if it corrected itself.

It got past the Getting Device Ready page and came up in the console to where I could see the logon.  At this point, I should not be able to see the console because the GRID driver takes over.  I would need to RDP into the VM to check on things.

RDP'd into the device and it was non-functional.  Could not launch the Nvidia Control panel.

Have you tried publishing just your OS layer and Platform layer?

 

One thing I failed to mention is that we do have to create a maintenance version on the PVS disk after its deployed from the app layering appliance and let it reboot once and then shut it down and promote that version to production.  From then on, we don't have the boot loop issue.  

 

Keep in mind, You'll have to cleanup the activation files that app layering creates otherwise OS/office won't activate.

 

https://support.citrix.com/article/CTX226062

 

TLDR: Remove the _done files from the C:\windows\setup\scripts\kmsdir and C:\windows\setup\scripts folders while you have the disk booted in maintenance mode to avoid the issue entirely.

 

We created a script that does this for us before shutting down.

[Riccardo Gasparini]

Hi,

The same exact problem happen with me.

The only thing that sometimes works for me is to do a platform layer with the driver NVidia and everything you need in it, finalize, add a version and play with it by adding the registry key and reinstalling the driver. This sometimes works and can result in a usable platform layer (but it's like flipping a coin).

 

In my case I can always see the GPU in the task manager but if the driver is not working correctly the control panel doesn't shows up at login and the right click menu on the desktop will hang a minute or two (probably because the button for the NVidia control panel is not present and this somehow create this problem).

Also, in my case, if it's not working in the build phase of the layer it will not work on the final image, no matter what.