Headless NVIDIA on CentOS/RHEL 6: VirtualGL 2.6 + TurboVNC 2.2 (Working Guide)

Headless NVIDIA + VirtualGL + TurboVNC on CentOS/RHEL 6 (EOL): A Working Guide

Why this post? CentOS/RHEL 6 are end-of-life, so current docs often break. This guide shows a known-working setup for running OpenGL apps over the network with VirtualGL 2.6.x and TurboVNC 2.2.x on headless NVIDIA GPUs (Tesla-era), including the critical NV-GLX / GLXCreateContext fixes.

Versions that still work on CentOS 6

• VirtualGL: 2.6.x (e.g., 2.6.5-20201117) — 3.x requires newer glibc.
• TurboVNC: 2.2.x (e.g., 2.2.90-20211222) — 3.x requires newer glibc and libXfont2.
• NVIDIA driver: 390.xx is a practical ceiling for many Fermi/Tesla cards.

Enable the VirtualGL & TurboVNC repos

Create two repo files (adjust baseurls if yours differ):

[root@host ~]# cat >/etc/yum.repos.d/VirtualGL.repo <<'EOF'
[VirtualGL]
name=VirtualGL
baseurl=https://virtualgl.org/pmwiki/uploads/Downloads/CentOS/6/x86_64
enabled=1
gpgcheck=0
EOF

[root@host ~]# cat >/etc/yum.repos.d/TurboVNC.repo <<'EOF'
[TurboVNC]
name=TurboVNC
baseurl=https://turbovnc.org/pmwiki/uploads/Downloads/CentOS/6/x86_64
enabled=1
gpgcheck=0
EOF

Install specific (compatible) versions

# VirtualGL 2.6.x (works on CentOS 6)
yum install -y VirtualGL-2.6.5-20201117

# TurboVNC 2.2.x (newer 3.x requires GLIBC_2.16+ and libXfont2)
yum install -y turbovnc-2.2.90-20211222

# 3D + fonts + GLX utilities
yum install -y xorg-x11-server-Xorg xorg-x11-xauth xorg-x11-xinit glx-utils \
               xorg-x11-fonts-misc xorg-x11-fonts-Type1 mesa-libGL mesa-dri-drivers

Configure a headless NVIDIA X server on :0

1. Find Bus IDs:

   nvidia-smi -pm 1
   nvidia-smi -q | egrep -i 'Bus Id|Bus'
   # Example GPU0 = 0000:02:00.0 → BusID "PCI:2:0:0"

2. Write /etc/X11/xorg.conf (headless NoScanout):

   cat >/etc/X11/xorg.conf <<'EOF'
   Section "ServerFlags"
       Option "DontVTSwitch" "true"
       Option "AllowEmptyInitialConfiguration" "true"
   EndSection
   
   Section "Device"
       Identifier  "GPU0"
       Driver      "nvidia"
       BusID       "PCI:2:0:0"
       Option      "UseDisplayDevice" "None"
       Option      "AllowEmptyInitialConfiguration" "true"
   EndSection
   
   Section "Monitor"
       Identifier  "DummyMon0"
       HorizSync   28.0-80.0
       VertRefresh 48.0-75.0
   EndSection
   
   Section "Screen"
       Identifier "Screen0"
       Device     "GPU0"
       Monitor    "DummyMon0"
       DefaultDepth 24
       SubSection "Display"
           Depth 24
           Virtual 1920 1080
       EndSubSection
   EndSection
   
   Section "Files"
       ModulePath "/usr/lib64/xorg/modules"
   EndSection
   
   Section "Module"
       Load "glx"
   EndSection
   
   Section "ServerLayout"
       Identifier "Layout0"
       Screen  "Screen0"
   EndSection
   
   Section "DRI"
       Mode 0666
   EndSection
   EOF

3. Make sure X loads NVIDIA’s GLX (not Mesa’s):

   ls -l /usr/lib64/xorg/modules/extensions/libglx.so
   # If it's tiny (~280 KB), repoint to NVIDIA (~14 MB):
   mv /usr/lib64/xorg/modules/extensions/libglx.so \
      /usr/lib64/xorg/modules/extensions/libglx.so.mesa
   ln -sf /usr/lib64/xorg/modules/extensions/libglx.so.390.30 \
         /usr/lib64/xorg/modules/extensions/libglx.so

4. Start the headless X and allow local clients:

   /usr/bin/X :0 -noreset -nolisten tcp -logfile /var/log/Xorg.0.log & sleep 5
   DISPLAY=:0 xhost +LOCAL:

VirtualGL + TurboVNC usage

# One-time server config (answers: enable, allow users, etc.)
/opt/VirtualGL/bin/vglserver_config

# Start a viewer desktop (TurboVNC)
/opt/TurboVNC/bin/vncserver    # connect with a VNC viewer

# In the VNC terminal (bash)
export VGL_DISPLAY=:0
vglrun -d :0 glxinfo | egrep -i 'opengl vendor|renderer|version'
vglrun -d :0 glxgears

Expect: OpenGL vendor = NVIDIA Corporation; renderer = Tesla M2090/…; gears window smooth.

Start at boot (SysV init)

Append to /etc/rc.d/rc.local and ensure it’s executable:

cat >>/etc/rc.d/rc.local <<'EOBOOT'
# --- Headless Xorg for VirtualGL on :0 ---
if ! pgrep -f "X .*:0" >/dev/null 2>&1 ; then
  /usr/bin/X :0 -noreset -nolisten tcp -logfile /var/log/Xorg.0.log &
  sleep 5
  DISPLAY=:0 /usr/bin/xhost +LOCAL:
fi
# --- end ---
EOBOOT
chmod +x /etc/rc.d/rc.local

Troubleshooting

• BadWindow (NV-GLX) ⇒ X on :0 is missing NVIDIA GLX. Recheck the libglx.so symlink and Load "glx" section; restart X and look for “NVIDIA GLX Module …” in /var/log/Xorg.0.log.
• X_GLXCreateContext BadValue (plain glxgears) ⇒ you’re trying direct GLX into a forwarded display. Use vglrun -d :0 ... or LIBGL_ALWAYS_INDIRECT=1 glxgears.
• Fonts warnings ⇒ install xorg-x11-fonts-* packages above.
• csh users ⇒ use setenv VGL_DISPLAY :0 instead of export.

Note: On CentOS/RHEL 6 you cannot install VirtualGL 3.x or TurboVNC 3.x due to glibc and libXfont2 requirements. Stick to the versions shown here.

💡 Before you begin: If your CentOS 6 system no longer updates or gives certificate errors, follow this CentOS 6 Vault repo & yum certificate fix guide first to restore working repositories.