How to disable the VirtualGL on Linux?

[aJokvaKo]

Hello,

I’m looking for the method to disable the VirtualGL of the NoMachine on ubuntu 16.04.3.

A few days ago, I installed the NoMachine version 5.3.10-6 on ubuntu 16.04.3 and enabled the VirtualGL following this instruction.

https://www.nomachine.com/AR11K00737

However, in my ubuntu server , the VirtualGL made some troubles. So, I’m looking for the method to disable the VirtualGL.

Please tell me.

 

Client OS: macOS 10.12.6

Client Side NoMachine : NoMachine version 5.3.10(free version)

Server OS: Ubuntu 16.04.3

Sever Desktop: Xfce 4.12.2

Server Side NoMachine:  NoMachine version 5.3.10-6(free version)

Sever GPUs: Quadro K620 & Tesla K80

* This server doesn’t have the display. (Headless server)

** I’m using the Nouveau display driver for Quadro K620 and the Nvidia official proprietary driver for Tesla K80, because the Linux Kernel 4.10 conflicts the Nvidia official proprietary display driver for Quadro K620. So, the process of VirtualGL runs on  Tesla K80. However, this makes a trouble: When I use the Tesla K80 for the heavy GPGPU program, the NoMachine become very slow.

This is the output of “nvidia-smai”

+—————————————————————————–+

| NVIDIA-SMI 384.69                 Driver Version: 384.69                    |

|——————————-+———————-+———————-+

| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |

| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |

|===============================+======================+======================|

|   0  Quadro K620         Off  | 00000000:02:00.0 Off |                  N/A |

| 34%   39C    P0     1W /  30W |     11MiB /  1998MiB |      0%      Default |

+——————————-+———————-+———————-+

|   1  Tesla K80           Off  | 00000000:84:00.0 Off |                    0 |

| N/A   41C    P0    59W / 149W |     96MiB / 11439MiB |      1%      Default |

+——————————-+———————-+———————-+

|   2  Tesla K80           Off  | 00000000:85:00.0 Off |                    0 |

| N/A   35C    P0    71W / 149W |     11MiB / 11439MiB |      0%      Default |

+——————————-+———————-+———————-+

 

+—————————————————————————–+

| Processes:                                                       GPU Memory |

|  GPU       PID  Type  Process name                               Usage      |

|=============================================================================|

|    1      3073    C   /usr/NX/bin/nxnode.bin                          85MiB |

+—————————————————————————–+

*** I tried the re-install of NoMachine on this server. (remove /usr/NX & re-install  NoMachine), but the VirtualGL proess( /usr/NX/bin/nxnode.bin) runs in the new NoMachine.

 

**** I tried to stop the VirtualGL from “/usr/NX/scripts/vgl/vglserver_config”.  I executed these commands,

1, sudo systemctl stop lightdm

2, sudo rmmd nvidia_uvm nvidia_drm nvidia_modeset nvidia

3, sudo /usr/NX/scripts/vgl/vglserver_config

Then,   typed “2” to executed “Unconfigure server for use with VirtualGL”.

4, sudo reboot

However, I couldn’t disable the VirtualGL.

 

[graywolf]

Did you turn off VirtualGL in node.cfg:

#
# Enable or disable loading VirtualGL libraries when starting virtual
# desktops on Linux.
#
# 1: Enabled. This make OpenGL applications able to use server side
#             graphics hardware.
#
# 0: Disabled. VirtualGL libraries are not loaded.
#
#EnableVirtualGLSupport 0

May I ask you which kind of troubles you got with VirtualGL?

[aJokvaKo]

Thank you for your reply.

> Did you turn off VirtualGL in node.cfg:

Yes, I did. However, VirtualGL is still running.

> May I ask you which kind of troubles you got with VirtualGL?

Currently, VirtualGL is running on Tesla K80 in my environment because of the Nouveau display driver for Quadro K620 (and Linux Kernel 4.10). However, I want to use Tesla K80 only for GPGPU computing. Now, I’m using MATLAB and Tesla K80  to analyze the data.  The size of each data is Tens of GB and these data require the bandpass filtering. So, when I use Tesla K80 for GPGPU, the memory of Tesla K80 becomes full and the VirtualGL process becomes very slow. As a result, the NoMachine stops responding.

Therefore, I want to disable VirtualGL.

Of Course, If the Linux Kernel stops conflicting the Nvidia official proprietary display driver for Quadro K620, I want to use VirtualGL on the Quadro K620.

[graywolf]

VirtualGL is turned off. The nxnode.bin process is using the video card for other purposes, likely video encoding.

Try to turn it off. In node.cfg change this key to 0:

#
# Enable or disable use of the hardware encoder.
#
# 1: Enabled. Use the hardware encoder if supported by the graphics
#    card.
#
# 0: Disabled. Don't use the hardware encoder.
#
#EnableHardwareEncoding 1

Then restart server with sudo /usr/NX/bin/nxserver --restart

[aJokvaKo]

Dear graywolf

Changing “EnableHardwareEncoding” key to 0 and executing sudo /usr/NX/bin/nxserver --restart stopped  nxnode.bin process on Tesla K80.

As a result, when I use Tesla K80 for GPGPU computing, the NoMachine works fine!

Thank you.

 

 

[Author]

Posts