Mageia forum

by **blosomax** » Feb 5th, '26, 04:04

NVIDIA GPU will not power off after use (powertop 100%) — real fix

Hardware: NVIDIA GP108M / GeForce MX150
Driver: 580.xx (Mageia nonfree)
Symptom: GPU can never reach powertop 0.0% or wont reach it after nvidia-modprobe -u -c=0 is run.

Update:
After running a series of power tests, I have to admit I was surprised by the results. On my machine, there’s no measurable difference in power usage between the NVIDIA card being “suspended” (Powertop showing 0.0%) and it being fully active (Powertop showing 100%), suggesting the NVIDIA driver is already managing its own power-down behavior internally. So this post turns out to be largely irrelevant 😄

Using the mageia-prime package (https://wiki.mageia.org/en/Mageia-prime_for_Optimus) with GPU offloading achieves essentially the same outcome in terms of overall power draw:

Feel free to keep reading if you’re still interested.

Issue

We want the NVIDIA discrete graphics card to be completely powered off when not in use, but still be able to turn it on when needed for offloading — without rebooting.

This assumes the system is already set up for PRIME render offload (not using NVIDIA for the whole desktop).

I'm posting this as I have spent several days trying to get the nvidia card to power off like it used to in the bumblebee bbswitch days. Hoping this post may save others some time.

Nvidia hardware
My system uses: PCI Device NVIDIA Corporation GP108M [GeForce MX150]
It's possible this isn't an issue for later nvidia cards which might support 'ture off' suspend without this hak.

bbswitch: Doesn't work, it says the card is OFF, but looking at powertop it is still at 100%

All my code examples assume:
- nvidia gpu located at pci:0000:01:00.0 (you may need to change this)
- everything is run as root (you may need to put 'sudo' in front of all the commands if you don't 'su' into root)

My minimum working install was:

Code: Select all: rpm -qa | grep nvidia

The package mageia-prime (https://wiki.mageia.org/en/Mageia-prime_for_Optimus) should also work (Which I recommend you use. If I had seen it before I started down this path I would have. I praise the mageia team for building it!. Please post if this works with it and ask questions if it doesn't. I'm happy to install it and get this going with it .... spent too many days to try now :-P

).
Note: if you are using mageia-prime: At first the nvidia card will not be able to be turned off. But you can skip to 'mageia-prime users can start here.' (below) to truly turn the nvidia card off.

The symptom

After boot, the GPU can suspend correctly (assuming nvidia drivers weren't loaded at boot, for this setup that's as simple as not doing anything to load them.
If you are using mageia-prime Skip this bit I'm pretty sure you wont be able to get the power to 0% yet:

Note: Toggling bad -> good for the NVIDIA card in powertop 'Tunables' will do the same as the first line in this code, and may help you easily find the pci address of your card.

Code: Select all: echo "auto" > /sys/bus/pci/devices/0000:01:00.0/power/control cat /sys/bus/pci/devices/0000:01:00.0/power/runtime_status

Result: suspended (Great!)

Powertop shows: 0.0% PCI Device: NVIDIA (totally off!)

Then to start the nvidia driver:
(I'm pretty sure mageia-prime already does this at boot so skip this also)

Code: Select all: nvidia-modprobe -u -c=0

mageia-prime users can start here.

The GPU becomes permanently stuck is active (no suspend) mode:

Code: Select all: echo "auto" > /sys/bus/pci/devices/0000:01:00.0/power/control cat /sys/bus/pci/devices/0000:01:00.0/power/runtime_status

Result: active

Powertop shows: 100.0% PCI Device: NVIDIA (wasting power if we're not using it)

Other things I tried to get it working (I literally spent days researching ... going down 'wrong' rabbit holes ... giving up ... starting again):

The GPU will not power off again until reboot.

I believe this is a PCI driver binding problem.

When you run:

Code: Select all: nvidia-modprobe -u -c=0

udev binds the PCI device to the NVIDIA driver:

Code: Select all: /sys/bus/pci/devices/0000:01:00.0 → /sys/bus/pci/drivers/nvidia

As long as a driver is bound to the PCI device: Linux PCI runtime power management is disabled.

Unloading modules does not unbind the driver.

That is why all normal methods fail.

The actual fix (no reboot)

Check who owns the device:

Code: Select all: readlink /sys/bus/pci/devices/0000:01:00.0/driver

You will see it points to nvidia.

Now do this:

Code: Select all: echo 0000:01:00.0 > /sys/bus/pci/drivers/nvidia/unbind echo auto > /sys/bus/pci/devices/0000:01:00.0/power/control

Immediately:

Code: Select all: echo "auto" > /sys/bus/pci/devices/0000:01:00.0/power/control cat /sys/bus/pci/devices/0000:01:00.0/power/runtime_status

Result: suspended (Great!)
Powertop shows: 0.0% PCI Device: NVIDIA (totally off!)

No module unload. No remove/rescan. No reboot.

Simple on/off scripts
Make a file (to be run as root) EG:

Code: Select all: touch /usr/local/sbin/nvidia-suspend.sh chmod +x /usr/local/sbin/nvidia-suspend.sh

Contents [updated]:

Code: Select all: #!/bin/bash GPU="0000:01:00.0" if [ "${1}" == "status" ] then cat "/sys/bus/pci/devices/${GPU}/power/control" cat "/sys/bus/pci/devices/${GPU}/power/runtime_status" elif [ "${1}" == "on" ] then if [ -L "/sys/bus/pci/devices/$GPU/driver" ] then echo "Already on? Try nvidia-smi as user" elif ! lsmod|grep -q nvidia then nvidia-modprobe -u -c=0 else nvidia-modprobe -u -c=0 echo "$GPU" > /sys/bus/pci/drivers/nvidia/bind fi elif [ "${1}" == "off" ] then if [ -L "/sys/bus/pci/devices/$GPU/driver" ] then echo "$GPU" > /sys/bus/pci/drivers/nvidia/unbind sleep 1 fi echo "auto" > /sys/bus/pci/devices/$GPU/power/control else echo "usage: ${0##*/}: on|off|status" fi

Now you can run (as root):

Code: Select all: nvidia-suspend.sh status nvidia-suspend.sh on nvidia-suspend.sh off

You can check results using:

Code: Select all: powertop

And check the nvidia card is working as a normal user. EG

Code: Select all: nvidia-smi

or if you installed mageia-prime pakage:

Code: Select all: mageia-prime-offload-run glxspheres64

Important notes

You do not need:

by **morgano** » Feb 5th, '26, 22:07

Thank you for the update :-)

So basically all is working OK, following https://wiki.mageia.org/en/Mageia-prime_for_Optimus ?
If you think something should be corrected or added in that page, please do.

by **blosomax** » Feb 6th, '26, 01:30

Thanks @morgano. I don't have permissions to edit wiki pages. I'd probably only add a note about not trying to get the nvidia card in suspend unless you can actually measure a power drain. I'm so used to the bbswitch bumblebee way of things, I just assumed the card would be wasting power when it's on (like it used to). Other people may think like me so it might be worth noting.

by **morgano** » Feb 7th, '26, 15:55

OK I added a note at bottom of https://wiki.mageia.org/en/Mageia-prime ... d_Graphics, with link to here.

You both test and write very well - you are much welcome to help in the Wiki

If you like, ask for edit rights in doc or QA mail list!

Mageia forum

NVIDIA GPU will not power off (powertop 100%) [solved]

NVIDIA GPU will not power off (powertop 100%) [solved]

Re: NVIDIA GPU will not power off (powertop 100%) [solved]

Re: NVIDIA GPU will not power off (powertop 100%) [solved]

Re: NVIDIA GPU will not power off (powertop 100%) [solved]

Who is online