Given that an AI has already been consulted and cited on this thread, I guess it's fine that I post the result of feeding all the above (attached log's content and Claude's chat included) to my special blend of ChatGPT 5.2 Plus, which follows. It's verbose and detailed. Is it accurate and/or useful? That, I don't know, but I hope it helps.
Diagnosis
- The Xorg log shows the Intel HD 4600 as the active scanout device:
Screen "intel"(0) is created, NVIDIA is configured as an Inactive device, and the NVIDIA Xorg driver is unloaded during startup. This matches a muxless/hybrid laptop where the internal panel (and often the external ports) are physically wired to the iGPU.
xrandr --listproviders showing only Intel is consistent with the above: only the GPU that Xorg actually wires up as a provider/sink in the running server will appear.- PRIME-style offloading on FreeBSD is described in the FreeBSD Handbook (X11/Graphics). When supported, it is typically exercised via
DRI_PRIME=1. See also Xorg(1)() for X server configuration mechanisms.
- The DRM kernel module ports (for example graphics/nvidia-drm-66-kmod) are tied to the current production NVIDIA branch and depend on x11/nvidia-kmod (current), not the legacy 470 kmod flavor. As a result, legacy 470 does not pair with these DRM-kmod ports.
- x11/nvidia-hybrid-graphics is deleted (expired as obsolete), so it cannot be used as an automation layer.
Net effect: “Disable Intel in xorg.conf and run NVIDIA-only” only works if at least one display output is physically wired to NVIDIA. If all outputs are iGPU-wired, forcing NVIDIA as the sole Xorg GPU usually produces “no screens” or a black display.
Corrective actions
1) Stop fighting autoconfiguration: remove the monolithic xorg.conf
HAZARD: This change can prevent the GUI from starting. Scope: Xorg display startup only. Keep a console login available.
The Handbook recommends avoiding a full
/etc/xorg.conf unless automatic configuration fails, and using
/usr/local/etc/X11/xorg.conf.d/ for targeted overrides. Xorg also supports configuration via xorg.conf.d directories; see
Xorg(1)().
Code:
# cp -a /etc/xorg.conf /etc/xorg.conf.bak
# mv /etc/xorg.conf /etc/xorg.conf.disabled
Verification:
2) Fix Vulkan selection/crashes first (common cause: wrong ICD or out-of-sync stack)
A segfault in
vkcube does not prove offload is working; it often indicates a loader/ICD mismatch or a driver bug. Use
graphics/vulkan-tools to see what Vulkan discovers and to control ICD selection.
Install tools and inspect devices:
Code:
# pkg install graphics/vulkan-tools
$ vulkaninfo --summary
List installed ICD manifests:
Code:
$ ls -1 /usr/local/share/vulkan/icd.d
Force one ICD for a test run (replace with the actual filename found above):
Code:
$ env VK_ICD_FILENAMES=/usr/local/share/vulkan/icd.d/<icd>.json vkcube
Keep these tests unprivileged (avoid running under doas/sudo) so environment variables are not filtered by “secure execution” behavior.
Optional loader debugging:
Code:
$ env VK_LOADER_DEBUG=all vkcube 2>&1 | sed -n '1,120p'
3) Use NVIDIA for OpenGL rendering without NVIDIA scanout (VirtualGL + a second X server)
HAZARD: Starting a second X server can disrupt input/VT handling on some setups. Scope: X session/virtual terminals on the host. Prefer testing from a local console and keep a recovery path.
This is the standard FreeBSD workaround when Intel is the only scanout path: keep the Intel X server for display, start a second (headless) X server bound to NVIDIA, and redirect OpenGL to it.
Install VirtualGL:
Code:
# pkg install x11/virtualgl
Find the NVIDIA BusID:
Code:
# pciconf -lv | grep -B3 -A6 -i nvidia
Create
/usr/local/etc/X11/xorg.conf.nv (adjust
BusID to match
pciconf):
Code:
Section "ServerLayout"
Identifier "nvidia-headless"
Screen 0 "Screen0"
EndSection
Section "Device"
Identifier "NVIDIAGPU"
Driver "nvidia"
BusID "PCI:1:0:0"
Option "AllowEmptyInitialConfiguration" "True"
EndSection
Section "Screen"
Identifier "Screen0"
Device "NVIDIAGPU"
EndSection
Start a second X server on display :8 (see
Xorg(1)() for options like
-sharevts):
Code:
$ X :8 -config /usr/local/etc/X11/xorg.conf.nv -sharevts -noreset -nolisten tcp &
Run OpenGL apps on NVIDIA and display them on the Intel desktop:
Code:
$ vglrun -d :8 glxinfo -B
$ vglrun -d :8 glxgears
Verification (expect NVIDIA as the OpenGL renderer inside the
vglrun output):
Code:
$ vglrun -d :8 glxinfo -B | egrep 'OpenGL vendor|OpenGL renderer'
Limitations:
- This targets OpenGL/GLX offload. It does not provide general Vulkan PRIME presentation to the Intel scanout.
4) NVIDIA-only Xorg (works only if a port is NVIDIA-wired)
HAZARD: This can produce a black screen or “no screens found.” Scope: display output availability. Test on an alternate display number; do not overwrite the working configuration.
Test without replacing the working Intel configuration:
Code:
$ X :1 -config /usr/local/etc/X11/xorg.conf.nvidia-only -retro -nolisten tcp
If no physical output is NVIDIA-wired, this will fail or start without any visible display.
Notes on nvidia-drm-kmod and legacy 470
- The graphics/nvidia-drm-kmod family is built around the current NVIDIA DRM module packaging and expects the current kmod stack (for example graphics/nvidia-drm-66-kmod for 580.x). These do not pair with the legacy 470 kmod flavor.
- Because of that mismatch, common “Optimus/PRIME” guidance that relies on NVIDIA DRM/KMS will not apply to a legacy 470 installation on FreeBSD 15.
