Configuration of PRIME

ECC83 · 2022-12-11 15:12:53

Hello!
On my system with hybrid graphics (Intel 9th Gen CPU + Nvidia RTX, lts kernel and proprietary drivers for Nvidia GPU) Nvidia GPU never goes into D3 power state. PRIME has been configured according to ArchWiki so the lowest power state is expected when the GPU is not in use.
I've figured out that the Nvidia GPU (instead of the Intel one) is always used by the Xorg process.
Could anybody help me with proper configuration please?

seth · 2022-12-11 15:28:01

Please post your xorg log, https://wiki.archlinux.org/title/Xorg#General

ECC83 · 2022-12-11 17:34:47

seth wrote:

Please post your xorg log, https://wiki.archlinux.org/title/Xorg#General

Here it is: https://termbin.com/qf0t

Last edited by ECC83 (2022-12-11 17:54:52)

seth · 2022-12-11 19:59:04

You're running on the intel chip and there's no output attached to the nvidia one.

systool -vm nvidia # to see whether NVreg_DynamicPowerManagement was applied (eg. if you've the module in the initramfs)

Did you monitor the power/control value?

cat /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/power/control

ECC83 · 2022-12-11 21:13:20

seth wrote:

Did you monitor the power/control value?
cat /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/power/control

The output of this ia "auto".
But the cat /sys/class/drm/card1/device/power_state shows D0. And the same is for card0.

seth · 2022-12-11 21:33:26

Did you check the condition of NVreg_DynamicPowerManagement?
(Also inspect the system journal for related errors)

ECC83 · 2022-12-11 21:58:39

I've set it as "0x02" via config file.

seth · 2022-12-11 22:11:02

I asked you to look whether that configuration is actually applied.
And in doubt, pass it to the kernel commandline.

ECC83 · 2022-12-11 22:14:01

Ok then, could you tell me please how can I check it?

seth · 2022-12-11 22:19:29

https://bbs.archlinux.org/viewtopic.php … 1#p2073221

ECC83 · 2022-12-11 22:32:06

The output of systool -vm nvidia is:
Module = "nvidia"

Attributes:
coresize = "55250944"
initsize = "0"
initstate = "live"
refcnt = "7681"
srcversion = "16DD2AB6F8E16BF9E137E59"
taint = "POE"
uevent = <store method only>
version = "520.56.06"

Sections:

seth · 2022-12-12 08:18:13

Please use code tags, https://bbs.archlinux.org/help.php#bbcode

Either the parameter isn't set or the module principally doesn't report parameters.
Please post your system journal and

seth wrote:

in doubt, pass it to the kernel commandline

nvidia.NVreg_DynamicPowerManagement=0x02

https://wiki.archlinux.org/title/Kernel_parameters

ECC83 · 2022-12-14 16:57:16

Sorry for the delay with answer.

Here is the output of # systool -v -m nvidia .
Here is the output of # journalctl -b | grep nvidia .

seth wrote:

in doubt, pass it to the kernel commandline

Also I dont know If I passed the parameter correctly. I've tried it with sysctl and with modprobe .

seth · 2022-12-14 17:08:59

"system journal" - not "grep"
You want to add "nvidia.NVreg_DynamicPowerManagement=0x02" to the kernel commandline to make sure it's applied, sysctl won't work, modprobe.d might, modprobe (likely) won't (because you may reasonably fail to unload the module)

ECC83 · 2022-12-14 17:22:39

Here is # journalctl -b.

seth wrote:

You want to add "nvidia.NVreg_DynamicPowerManagement=0x02" to the kernel commandline to make sure it's applied, sysctl won't work, modprobe.d might, modprobe (likely) won't (because you may reasonably fail to unload the module)

Could you please explain then how to do it?

seth · 2022-12-14 17:32:36

seth in #12 wrote:

https://wiki.archlinux.org/title/Kernel_parameters

Edit: from the journal the nvidia module looks very much to be in the initramfs, so this is relevant.

Last edited by seth (2022-12-14 17:36:39)

ECC83 · 2022-12-14 17:43:27

seth wrote:

seth in #12 wrote:
https://wiki.archlinux.org/title/Kernel_parameters
Edit: from the journal the nvidia module looks very much to be in the initramfs, so this is relevant.

I'e read this. It looks like I don't have solid understanding of the process anyway. For example I have no idea why the config file from /etc/modprobe.d/ has no effect.
Conserning initramfs: I've added to /etc/mkinitcpio.conf MODULES=(nvidia nvidia_modeset nvidia_uvm nvidia_drm).

Last edited by ECC83 (2022-12-14 17:48:59)

seth · 2022-12-14 20:07:50

I've added to /etc/mkinitcpio.conf MODULES=(nvidia nvidia_modeset nvidia_uvm nvidia_drm).

Which is why the settings in modprobe.d won't apply until you recreate the initramfs.
That's why I repeatedly told you to utilize the kernel commandline, because it makes the exact want and point when the module is loaded irrelevant.

ECC83 · 2022-12-15 15:25:53

seth wrote:

I've added to /etc/mkinitcpio.conf MODULES=(nvidia nvidia_modeset nvidia_uvm nvidia_drm).
Which is why the settings in modprobe.d won't apply until you recreate the initramfs.
That's why I repeatedly told you to utilize the kernel commandline, because it makes the exact want and point when the module is loaded irrelevant.

Actually I've recreated it right after editing of mkinitcpio.conf by running # mkinitcpio -P . All the data I posted earlier here by your requests were provided after that.

Last edited by ECC83 (2022-12-15 15:26:41)

seth · 2022-12-15 15:49:21

cat /proc/driver/nvidia/gpus/0000:01:00.0/power

ECC83 · 2022-12-15 18:13:02

seth wrote:

cat /proc/driver/nvidia/gpus/0000:01:00.0/power

Here it is.
Also cat /sys/class/drm/card1/device/power_state gives D0.

seth · 2022-12-15 20:04:42

Runtime D3 status:          Enabled (fine-grained)

So the parameter applied, what does nvidia-smi actually report about the consumers?
(It is normal that Xorg "uses" the nvidia GPU, after all it's available for prime, but it should™ not tax it at all unless you're prime-running something or maybe nvidia-settings or nvidia-smi excessively poll the GPU)

ECC83 · 2022-12-16 13:07:36

seth wrote:

Runtime D3 status:          Enabled (fine-grained)
So the parameter applied, what does nvidia-smi actually report about the consumers?
(It is normal that Xorg "uses" the nvidia GPU, after all it's available for prime, but it should™ not tax it at all unless you're prime-running something or maybe nvidia-settings or nvidia-smi excessively poll the GPU)

The nvidia-smi reports this. I know it's far from maximum power consumption. But according to the wiki: "it is possible to fully power down the GPU when not in use". Which is not the case. That's why I think maybe I missed somesthing.

V1del · 2022-12-16 13:10:48

nvidia-smi will implicitly poke the GPU afaik and in general sensor listings on laptops are unreliable because normally not available distinctly for each piece of HW. What is your power draw in powertop? If the overall draw is within reason then this isn't further problematic. If the 7W you see there is the whole consumption then the GPU is basically turned off and the 7W just a general report.

seth · 2022-12-16 13:43:04

You could also try to deactivate the persistenced and

Dec 14 20:16:01 ECC83 systemd-udevd[440]: nvidia_drm: Process '/bin/chgrp wheel /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:01 ECC83 systemd-udevd[440]: nvidia_drm: Process '/bin/chmod g+w /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:01 ECC83 systemd-udevd[324]: nvidia_modeset: Process '/bin/chgrp wheel /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:01 ECC83 systemd-udevd[324]: nvidia_modeset: Process '/bin/chmod g+w /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:01 ECC83 systemd-udevd[442]: nvidia_uvm: Process '/bin/chgrp wheel /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:01 ECC83 systemd-udevd[442]: nvidia_uvm: Process '/bin/chmod g+w /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:01 ECC83 systemd-udevd[380]: nvidia: Process '/bin/chgrp wheel /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:01 ECC83 systemd-udevd[380]: nvidia: Process '/bin/chmod g+w /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
…
Dec 14 20:16:02 ECC83 systemd-udevd[334]: nvidia-nvswitch: Process '/bin/chgrp wheel /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:02 ECC83 systemd-udevd[334]: nvidia-nvswitch: Process '/bin/chmod g+w /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:02 ECC83 systemd-udevd[440]: nvidia: Process '/bin/chgrp wheel /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.
Dec 14 20:16:02 ECC83 systemd-udevd[440]: nvidia: Process '/bin/chmod g+w /sys/class/backlight/intel_backlight/brightness' failed with exit code 1.

whatever that mess is.

Arch Linux

#1 2022-12-11 15:12:53

Configuration of PRIME

#2 2022-12-11 15:28:01

Re: Configuration of PRIME

#3 2022-12-11 17:34:47

Re: Configuration of PRIME

#4 2022-12-11 19:59:04

Re: Configuration of PRIME

#5 2022-12-11 21:13:20

Re: Configuration of PRIME

#6 2022-12-11 21:33:26

Re: Configuration of PRIME

#7 2022-12-11 21:58:39

Re: Configuration of PRIME

#8 2022-12-11 22:11:02

Re: Configuration of PRIME

#9 2022-12-11 22:14:01

Re: Configuration of PRIME

#10 2022-12-11 22:19:29

Re: Configuration of PRIME

#11 2022-12-11 22:32:06

Re: Configuration of PRIME

#12 2022-12-12 08:18:13

Re: Configuration of PRIME

#13 2022-12-14 16:57:16

Re: Configuration of PRIME

#14 2022-12-14 17:08:59

Re: Configuration of PRIME

#15 2022-12-14 17:22:39

Re: Configuration of PRIME

#16 2022-12-14 17:32:36

Re: Configuration of PRIME

#17 2022-12-14 17:43:27

Re: Configuration of PRIME

#18 2022-12-14 20:07:50

Re: Configuration of PRIME

#19 2022-12-15 15:25:53

Re: Configuration of PRIME

#20 2022-12-15 15:49:21

Re: Configuration of PRIME

#21 2022-12-15 18:13:02

Re: Configuration of PRIME

#22 2022-12-15 20:04:42

Re: Configuration of PRIME

#23 2022-12-16 13:07:36

Re: Configuration of PRIME

#24 2022-12-16 13:10:48

Re: Configuration of PRIME

#25 2022-12-16 13:43:04

Re: Configuration of PRIME

Board footer