NM acceleration possible when it creates its own display

[pmanousos]

Greetings – using NM 7.6.2 on RHEL 7 server side, Win 10 Enterprise Client side.

We notice when a desktop manager is running on RHEL7 (not initiated by NM) we see their is access to full hardware acceleration (example glxinfo run in a terminal window from a NM session on the client side shows all extensions being pushed through to the client).

However, in the absence of a desktop manager running, NM will ask to create a display (which we accept) and despite it starting its own session of the desktop manager (in our case Gnome <the gdm>) we see there is no longer full hardware acceleration being pushed through to the client side.   A terminal window from a NM session on the client side (after typing the command glxinfo) shows only SOME of the extensions being pushed through.

We verified that “Use acceleration for display processing” is checked (enabled) in the Server Settings gui on the server side.

We were not able to find this particular issue specified in the forum (there were related topics) but none seemingly specific for when NM creates a display for the client side when there is no desktop manager running already on the server side.

We apologize if this IS covered in the form so could you point us to the right threat of if this is not in the form could you answer for us is this a limitation of NM or is there some configuration we could try to over come this limitation?

[Britgirl]

This is a functionality supported in, for example, NoMachine Workstation, of the Terminal Server range. Take a look at this page: https://www.nomachine.com/terminal-server.

[pmanousos]

Thank you for assisting.  Per your implied recommendation we uninstalled the workstation eval version and installed terminal server (7.8.2_1_x86_64.rpm) but this still gives the same results and wanted to bring this to your attention.  We may not have been detailed enough in our original response, so please see details below because we wish to purchase your product but only if we can get it working properly.

Scenario1 (gnome desktop manager running prior to !M server being launched) — works as expected

1. On our RHEL 7 server, prior to !M running in the background,  as root we run   systemctl start gdm

2. On our RHEL 7 server, start !M  /usr/NX/bin/nxserver –restart 3. On our client (Windows 10 laptop), start nxplayer.exe,  connect to server successfully using “user1” non-root credentials, select “Physical display, gdm, Linux desktop on :0”

3. Result – we see the desktop of the remote server, a critical application we launch runs perfectly,  glxinfo | grep imaging shows “GL_ARB_imaging” <—-important!

 

Scenario2 (gnome desktop manager initiated by !M connection to the server) – does NOT work as expected

1. On RHEL 7 server, systemctl stop gdm

2. On RHEL 7 server, restart !M   /usr/NX/bin/nxserver –restart

3. On our client (Windows 10 laptop), start nxplayer.exe,  connect to server successfully using “user1” non-root credentials, select “Create a new virtual desktop”

4. Result – we see a scaled-down version of the desktop of the remote server, a critical application we launch does not render certain graphics,  glxinfo | grep imaging returns nothing <—-important!

 

In both scenarios, “Use hardware encoding” and “Use X11 vector graphics mode in virtual sessions” are enabled.

So even with Terminal Server version of !M, we still get the results as originally described which is as follows:
“We notice when a desktop manager is running on RHEL7 (not initiated by NM) we see access to full hardware acceleration (glxinfo run in a terminal window from a NM session on the client side shows all extensions being pushed through to the client).   However, in the absence of a desktop manager running, NM will ask to create a display (which we accept) and despite it starting its own session of the desktop manager (in our case Gnome <the gdm>) there no full hardware acceleration being pushed through to the client side — a terminal window from a NM session on the client side (after typing the command glxinfo) shows only SOME of the extensions being pushed through.”

Can you offer us additional advice to get around this issue?   We are not sure why some of these extensions of glx are not being pushed through when running !M in “virtual mode”

[Britgirl]

Hi,

I wrongly assumed you had installed the free version on the remote host you are accessing hence the suggestion to install Workstation or another product from the Terminal Server range. You didn’t need to uninstall NoMachine Workstation and then install Terminal Server, so sorry for the confusion. They both support HW acceleration in NM virtual desktop sessions provided you enable VirtualGL on the server where you have installed Workstation/Terminal Server. We checked that the extension that you highlight as important is present in VirtualGL.

Please verify that you do have VirtualGL installed and that is enabled. Prequisites are that the host machine is a physical computer and has a hardware-accelerated 3D video card, and that Xorg is up and running to use the 3D video card.

You can follow the instructions here to enable VirtualGL:

How to enable VirtualGL support on Linux in NoMachine v. 6.2 or later
https://knowledgebase.nomachine.com/AR05P00982

[pmanousos]

Greetings and thank you for this lead.   Per instructions on https://knowledgebase.nomachine.com/AR05P00982 we performed the command sudo /etc/NX/nxserver --virtualgl-install   and get  the following

NX> 900 Done. You must restart the display manager for the changes to take effect.

NX> 900 IMPORTANT NOTE: Your system uses modprobe.d to set device permissions.  You

NX> 900 must execute ‘rmmod nvidia_drm nvidia_modeset nvidia’ with the display manager

NX> 900 stopped in order for the new device permission settings to become effective.

But upon executing the recommended command (as root)   rmmod nvidia_drm nvidia_modeset nvidia  we run into a brick wall…

rmmod: ERROR: Module nvidia_drm is in use

rmmod: ERROR: Module nvidia_modset is not currently loaded

rmmod: ERROR: Module nvidida is not currently loaded

Can you offer any advice on how to proceed (“nvidia_drm” yielded seemingly no pertinent results on the forum)

[Britgirl]

Did you stop the display manager before executing ‘rmmod nvidia_drm nvidia_modeset nvidia’?

[pmanousos]

Yes – we verified there was no display manager running (we run gnome so there was no gdm or X server running in the background).

[pmanousos]

An update – while NX server was shutdown, and while there is not a desktop manager running (no gdm), and no X server running we re-ran the command as root /etc/NX/nxserver –virtualgl-install and got normal feedback with the same “you must execute rmmod nvidia_drm nvidia_modeset nvidia”  for setting to become effective.   And when we did then run that command as root we got slightly different error messages

rmmod: ERROR: Module nvidia_drm is not currently loaded

rmmod: ERROR: Module nvidia_modeset is not currently loaded

rmmod: ERROR: Module nvidia is in use by: nvidia_uvm

[Britgirl]

It doesn’t seem to be directly related to NoMachine (you say you shut down the NoMachine server).  You could try booting the host in runlevel 3. Alternatively, the nvidia forums might be able to provide some advice on those errors you are getting.

[pmanousos]

Hello again Britgirl.  And thank you so far for your help.

We were able to get around the rmmod issue  by running as root modprobe -r nvidia_uvm.   However – completing the instructions as listed on https://knowledgebase.nomachine.com/AR05P00982 did not give expected results.  Here are the details.

1. Completed instructions https://knowledgebase.nomachine.com/AR05P00982 with the following items of note

Because nvidia-xconfig –query-qpu-info shows 1 display device we consider our virtual server to have a “head” and therefore did not complete “headless Nvidia how to” instructions

Completed the configuration for the system to user virtualgl   by as root  executing /etc/NX/nxserver --virtualgl-install iand the resulting directive to  execute   rmmod nvidia_drm nvidia_modeset nvidia  (which we needed to complete using modprobe -r nvidia_uvm)
Ensured  NX was configured to run all  apps via virtualGL  by executing as root  /etc/NX/nxserver --virtualgl yes
Verified   EnableVirtualGLSupport 1     exists in /usr/NX/etc/node.cfg
Added both the following in /usr/NX/etc/node.cfg
DefaultDesktopCommand "/etc/gdm/Xsession 'env XDG_SESSION_TYPE=x11 /usr/NX/scripts/vgl/vglrun gnome-session --session=gnome'"
CommandStartGnome "/etc/gdm/Xsession 'env XDG_SESSION_TYPE=x11 /usr/NX/scripts/vgl/vglrun gnome-session --session=gnome"

2. We are not sure if we need to /etc/NX/nxserver --virtualgl no  since we added those DesktopCommands, so we tried it both ways (/etc/NX/nxserver --virtualgl yes   and   no  )  and both /came up empty.  Here is what we mean by “came up empty”

Ensured gdm (gnome desktop or any X server was running)
Restart nxserver as root   /usr/NX/bin/nxserver --restart    to ensure any changes in config file were being used
On the client side launched NM on windows and connected to our server normally (using the user account credentials on the server
The client interface asked “Create New Virtual” or “Create New Custom” and we clicked “Create New Virtual” and after stepping through the options got a quick black screen in the interface then were bumped back to the “Create New Virtual” or “Create New Custom” options

So at this point we can confirm that VirtualGl is installed on our RHEL 7 system  running nxserver (specifially its VirtualGL-2.5.2-1.el7.x86_64) and that as best we can tell have configured our system and NX server to match the instructions provided on https://knowledgebase.nomachine.com/AR05P00982  but still are not successful in launching a virtual session.

Two last pieces of debugging information that might help you.

1. When we launch the gdm as root   systemctl start gdm   and start a NX session it works just fine  (NX Client allows us to use the session with the physical display).

2. But when gdm is not running in the background and we fail to start an NX session we note the following in /var/log/message (curious why it cant open a display because the nx.config seems to be setting a virtual one) but here is the output

Mar 22 15:17:23 lwthr3p systemd: Started Session 1096 of user fewx.

Mar 22 15:17:24 lwthr3p dbus[1677]: [system] Activating via systemd: service name=’org.bluez’ unit=’dbus-org.bluez.service’

Mar 22 15:17:24 lwthr3p com.redhat.imsettings: [ 1647976644.684352]: IMSettings-Daemon[36322]: INFO: Starting imsettings-daemon…

Mar 22 15:17:24 lwthr3p com.redhat.imsettings: [ 1647976644.684441]: IMSettings-Daemon[36322]: INFO:   [HOME=/home/fewx/.config/imsettings]

Mar 22 15:17:24 lwthr3p com.redhat.imsettings: [ 1647976644.684469]: IMSettings-Daemon[36322]: INFO:   [XINPUTRCDIR=/etc/X11/xinit/]

Mar 22 15:17:24 lwthr3p com.redhat.imsettings: [ 1647976644.684492]: IMSettings-Daemon[36322]: INFO:   [XINPUTDIR=/etc/X11/xinit/xinput.d/]

Mar 22 15:17:24 lwthr3p com.redhat.imsettings: [ 1647976644.684521]: IMSettings-Daemon[36322]: INFO:   [MODULEDIR=/usr/lib64/imsettings]

Mar 22 15:17:24 lwthr3p com.redhat.imsettings: [ 1647976644.684545]: IMSettings-Daemon[36322]: INFO:   [MODULES=gsettings]

Mar 22 15:17:30 lwthr3p gnome-session-binary[36577]: WARNING: software acceleration check failed: Child process exited with code 1

Mar 22 15:17:30 lwthr3p gnome-session: gnome-session-binary[36577]: WARNING: software acceleration check failed: Child process exited with code 1

Mar 22 15:17:30 lwthr3p journal: Cannot open display:

Mar 22 15:17:30 lwthr3p com.redhat.imsettings: [ 1647976650.749609]: IMSettings-Daemon[36322]: INFO: Release the ownership of com.redhat.imsettings

Mar 22 15:17:30 lwthr3p org.gtk.vfs.Daemon: A connection to the bus can’t be made

Mar 22 15:17:30 lwthr3p com.redhat.imsettings: Exiting…

Mar 22 15:17:30 lwthr3p com.redhat.imsettings: [ 1647976650.749789]: GLib-GIO[36322]: CRITICAL **: Error while sending AddMatch() message: The connection is closed

Mar 22 15:17:30 lwthr3p com.redhat.imsettings: [ 1647976650.749845]: GLib-GIO[36322]: CRITICAL **: Error while sending AddMatch() message: The connection is closed

Mar 22 15:17:30 lwthr3p com.redhat.imsettings: [ 1647976650.749919]: IMSettings-Daemon[36322]: INFO: Unloading imesttings module: gsettings

Mar 22 15:17:30 lwthr3p com.redhat.imsettings: [ 1647976650.749972]: IMSettings-Daemon[36322]: INFO: imsettings-daemon is shut down.

Mar 22 15:17:34 lwthr3p systemd-logind: Removed session 1096.

Mar 22 15:17:49 lwthr3p dbus[1677]: [system] Failed to activate service ‘org.bluez’: timed out

[pmanousos]

Britgirl – found the issue — it is we werent running the X server.   We were in the habit of having it only be launched by GDM that we didnt have it running after these changes.   So this works!!

We can launch multiple simultaneous independent sessions to the same server and they all push the needed extensions of glxinfo to the client.

Thank for your help and you can mark this issue as resolved.

[Britgirl]

Thanks for the additional info. We’re glad to know you resolved the problem because so far we hadn’t been unable to reproduce the behaviour, and I was about to suggest you try with a different desktop environment such as Mate or XFCE.

[Britgirl]

Just to answer your questions from your last reply.

We are not sure if we need to /etc/NX/nxserver --virtualgl no  since we added those DesktopCommands, so we tried it both ways (/etc/NX/nxserver --virtualgl yes   and   no

Use  /etc/NX/nxserver --virtualgl yes only. Setting it to ‘no’ would just disable VirtualGL

 

Added both the following in /usr/NX/etc/node.cfg
DefaultDesktopCommand “/etc/gdm/Xsession ‘env XDG_SESSION_TYPE=x11 /usr/NX/scripts/vgl/vglrun gnome-session –session=gnome'”
CommandStartGnome “/etc/gdm/Xsession ‘env XDG_SESSION_TYPE=x11 /usr/NX/scripts/vgl/vglrun gnome-session –session=gnome”

It’s not necessary to insert vglrun. You could try with

DefaultDesktopCommand "/bin/dbus-launch --exit-with-session gnome-session --session=gnome --disable-acceleration-check"

[Author]

Posts