Unable to change brightness while using iGPU only mode.

[RPochyly]

On my laptop, display brightness cannot be adjusted while the system is switched to integrated-GPU-only mode. By default the kernel only exposes one backlight interface in /sys/class/backlight, and it is not the device that actually drives the panel in iGPU mode. The AMD backlight device (amdgpu_bl1) is missing unless specific kernel parameters are passed, and even once it is present, its default brightness scale is non-linear, leaving the screen visibly dimmer at 100% than in hybrid/dGPU mode. I have also found kernel parameters which should fix all these issues. If this is not the right project for this issue, please redirect me elsewhere.

System information

• Laptop: ASUS TUF Gaming A16 (FA608WI)
• iGPU: AMD Radeon 890M
• dGPU: NVIDIA RTX 4070
• Distribution: CachyOS
• Kernel: 7.0.11-1-cachyos
• AMD driver: mesa 2:26.1.2-1
• NVIDIA driver: nvidia-open-dkms 610.43.02-3

The laptop supports three graphics modes, and backlight behavior is inconsistent across some of them. Hybrid and dGPU mode work okay, however when switching to iGPU only mode (integrated), the backlight becomes stuck on 100%. Additionally 100% on iGPU is not equal to 100% on Hybrid or dGPU modes.

1. No working backlight device in iGPU-only mode

With a default kernel command line, /sys/class/backlight does not expose the device needed to control brightness in iGPU-only mode. The desktop brightness controls (KDE Plasma slider, brightness keys) do nothing. The correct device, amdgpu_bl1, is not shown in /sys/class/backlight.

2. Non-linear amdgpu brightness scale

The panel is noticeably dimmer at 100% brightness in iGPU-only mode than it is in hybrid or dGPU mode. The cause is the default amdgpu brightness scale being non-linear:

$ cat /sys/class/backlight/amdgpu_bl1/scale
non-linear

In hybrid/dGPU mode the panel reaches full brightness as expected, so this is specific to the amdgpu backlight path.

Steps to reproduce

1. Boot with a default kernel command line.
2. Switch the laptop to iGPU-only mode.
3. Try to change brightness via the Plasma slider, brightness keys, or by writing to /sys/class/backlight/*/brightness.
4. Observe that brightness does not change (no amdgpu_bl1 device present), or — once the device is forced — that 100% is dimmer than in hybrid/dGPU mode.

Expected behavior

Brightness should be adjustable in all three graphics modes, and 100% brightness should produce the same panel output in iGPU-only mode as it does in hybrid/dGPU mode.

Workarounds

Enable both backlight devices

Add the following kernel parameters to expose both backlight interfaces in /sys/class/backlight:

acpi_backlight=native nvidia_wmi_ec_backlight.force=1

After this, both devices are present:

$ ls /sys/class/backlight/
amdgpu_bl1  nvidia_wmi_ec_backlight

Brightness then has to be written to the device that matches the active mode:

• iGPU-only mode -> write to amdgpu_bl1
• Hybrid / dGPU mode -> write to nvidia_wmi_ec_backlight

Fix the non-linear scale (dim-at-100% issue)

Add this additional parameter so the amdgpu backlight uses a brightness curve that reaches full panel brightness:

amdgpu.dcdebugmask=0x40000

[NeroReflex]

Is there a reason for you to disable the dgpu?

[RPochyly]

Mostly for power saving. The dGPU is being woken by Discord and some other programs (mostly Electron apps I think) from D3cold.

I've heard about cardwire. However when I tried it for a few minutes, Steam couldn't be launched. I wanted to play Terraria which doesn't need the dGPU on as it's quite simple and 2D.

I think being able to fully disable the dGPU provides the most flexibility. I'm now using a script which can completely disable or enable the dGPU without rebooting using the WMI, but I'm not sure right now. If you want, I can check and provide you the script this evening.

[NeroReflex]

Please tell me the script isn't removing the nvidia from the bus....

Anyway the solution is have cardwire working.

Short-term you can use libnvidia-hide to hide the nvidia from electron-based applications (I can help with that).

Note that disabling the iGPU is not really suggested and is not something that will see time poured in....

[RPochyly]

I basically write 1 to /sys/devices/platform/asus-nb-wmi/dgpu_disable after unloading nvidia modules. To re-enable it, I write 0 to the same file and trigger a PCIe rescan using /sys/bus/pci/rescan. I tried to replicate what GHelper does to enable iGPU-only mode.

Please tell me the script isn't removing the nvidia from the bus....

Yes, it removes the nvidia from the bus. Is that actually a problem, though? It has worked great for me so far, so I'm curious whether it's just considered bad practice, or whether it could cause damage or anything similar.

Just for my own understanding, though: what makes a full dGPU disable worse than D3cold + cardwire? Intuitively, turning the dGPU off completely seems like the easier solution to implement, whereas intercepting and blocking calls to nvidia device sounds more complex. I assume there's a reason I'm not seeing. I'll also take a look at libnvidia-hide, thanks for the suggestion and your help!

[superm1]

Yes, it removes the nvidia from the bus. Is that actually a problem, though? It has worked great for me so far, so I'm curious whether it's just considered bad practice, or whether it could cause damage or anything similar.

Well as it pertains to this bug you have - amdgpu has no idea it's supposed to reprobe brightness. This is going to get even worse when https://lore.kernel.org/dri-devel/20260531114908.1693426-1-superm1@kernel.org/ is merged...

[NeroReflex]

I deprecated supergfxctl exactly because hiding the nvidia that way (masking it from the bus) caused problems to many users. It returns in a badly configured after sleep+resume cycle and on some laptops the nvidia doesn't even sleep while removed: it's unusable AND draining power.

The current cardwire solution is hugely better because it works with the driver to keep it at lowest power draw (effectively zero) instead of working against the driver. Hiding the gpu to software is because certain software doesn't know how to behave and opens the nvidia gpu even if there is no need. Currently nvidia has a bug (that 2 years ago promised to investigate) that loading vulkan to enumerate devices will wake the dgpu if its rendernode is not made inaccessible to software. Nothing we can do about that if not masking the dgpu to software that has no business opening it...

So as far as the current solution sounds more complicate it actually is way less convoluted that dancing with drivers masking, events not finiring properly and/or firmware-attributes like dgpu_disable that behaves differently for each laptop.

[RPochyly]

Okay, I get it now. It seems that cardwire is the most universal solution going forward. Thank you for the explanations!

I'll try to troubleshoot the Steam problems and open an issue on the cardwire repo if I manage to reproduce them.