How to switch off NVidia Optimus dGPU (Maxwell)?

james-h-arch · 2024-01-29 13:39:06

After a long time successfully using optimus-manager to control the discrete GPU in my NVidia Optimus laptop, I'm currently unable to switch the dGPU off.

With dGPU always on, the laptop runs hotter, noisier and with more battery consumption than before.

I monitor system temperature & fan speed when under minimal load. With dGPU properly powered off, I had a nice silent laptop with no fans. Now, the fans frequently kick in.

I suspect the underlying systems have moved on. With optimus-manager no-longer being maintained, its various power-management methods no-longer apply.

Things may possibly be further complicated by the dGPU being Maxwell architecture (NVidia Quadro M1200).

I've tried:

Disabling in BIOS: Not supported (it's a Dell Precision device)
Optimus-manager methods Nouveau, acpi: Both used to work, now neither work.
udev rules from the Arch wiki here: Removed dGPU from the pci bus okay. dGPU remained powered-on (based on temps, fans)

How can I switch off NVidia Optimus dGPU?

Last edited by james-h-arch (2024-01-29 15:59:23)

seth · 2024-01-29 16:26:34

You're not getting https://wiki.archlinux.org/title/PRIME# … Management with that chip, but it seems wrong for it to generate considerable heat while idling
Please check nvidia-smi for consumers and actual chip temperature and post your Please post your Xorg log, https://wiki.archlinux.org/title/Xorg#General

How exactly does, I guess, https://wiki.archlinux.org/title/Hybrid … _acpi_call not work? Does turn_off_gpu.sh fail for every bus?

james-h-arch · 2024-01-29 17:35:13

Thanks Seth, appreciate your advice.

Here's the various files:

With the acpi_call "not working", the call itself appears successful in that there's no errors and nvidia-smi can no longer see the dGPU. However the dGPU does not seem to be powered off.

nvidia-smi output after acpi_call

It's a bit tricky for me to fully assess whether dGPU is really on or off. I'm monitoring indirectly via system fan speeds. Fans remain on pretty much indefinitely several minutes after the acpi call. There's the occasional few mins when the fans go off. They switch on again a few mins later.

By contrast, under Win-10 the fans switch off and stay off soon after login (it's a dual-boot machine. I have disabled Win-10 fast start).

seth · 2024-01-29 21:40:53

The GPU is not in use by any process and hangs out at 42°C
And the updated smi status shows that the GPU gets powered down because

Trying \_SB.PCI0.PEG0.PEGP._OFF: works!

I don't think that the GPU is the source of your heat issues.
Run powertop and check for the biggest consumer (the heat ultimately comes from your battery) and of course see top for CPU load as well as https://wiki.archlinux.org/title/Lm_sensors

james-h-arch · 2024-01-31 13:37:19

Thanks for the pointer re. powertop.

Powertop installed, calibrated on battery power for couple of hours and two reports generated:

I don't know much about 'normal' laptop power consumption. The values reported here seem low to my uninformed eye (about 7 Watts in total).

I ran watch on CPU frequencies and sensors.

watch "grep \"^[c]pu MHz\" /proc/cpuinfo; sensors"

Values remained stable throughout the powertop test. Here's a typical snapshot:

CPU frequencies and sensor values

Just in case it would help, I then tried unplugging a USB dongle (used for a wireless mouse/keyboard) & disabling the onboard BT radio. There was no significant change to sensors. Fan speed remained at 2.5k rpm.

seth · 2024-01-31 14:32:29

So the main consumers are the audio chip (?) and radio - 7W doesn't sound excessive, though.
The only heat source there seems to be the nvme => check iotop (maybe one of the file indexers baloo/tracker/… goes wild)

Otherwise, does the system actually heat up (the overall temperatures look very uncritical) or is this just an issue w/ https://wiki.archlinux.org/title/Fan_speed_control ?

james-h-arch · 2024-01-31 17:00:39

Yes, the audio chip reported at 100% looks odd. From a spot of further research, it seems to be a feature and doesn't necessarily indicate excessive power draw. I'm reassured by the overall 7W draw.

With iotop, things didn't look too wild (although I'm unsure what good/bad looks like):

iotop examples

With the system heat, yes it feels a little warm to the touch. Not hot.

Thanks for the fan control link too. I spent some time a while ago looking at Dell laptop fan control options. I concluded at the time that the onboard hardware/BIOS was auto-handling fans, driven by temperatures. Also that it felt more sensible leaving Dell's system in place and not seeking to override it.

seth · 2024-01-31 21:42:07

There's barely any IO in that recording.

Since the GPU gets powered down by acpi_call and assuming the fans remain running regardless and despite the notebook not overly hot (did you wait for some further cooldown after the acpi_call deactivated the GPU?)
a) did you install a firmware (UEFI) update?
b) can you configure the fan profile/behavior in the UEFI?
c) ceterum censeo: is there a parallel windows installation?

What specific model is this?

james-h-arch · 2024-02-01 12:00:33

There's barely any IO in that recording.

Yes, I kept the machine very lightly loaded while running the fan tests.

Since the GPU gets powered down by acpi_call and assuming the fans remain running regardless and despite the notebook not overly hot (did you wait for some further cooldown after the acpi_call deactivated the GPU?)

Yes, fans remained running at 2.5k rpm after dGPU acpi call. I waited and monitored for about another 60 minutes.

a) did you install a firmware (UEFI) update?

Since buying the machine? Yes, a couple of times. OEM updates (Dell), nothing fancy.

Installed recently? No - although to be fair I cannot recall when I last updated the BIOS and whether that correlated with changing fan behaviour.

I've checked and the latest firmware available is v.1.35.0 , released 12 Dec 2023. I've installed v.1.33.0, released 07 Apr 2023.

The firmware release notes aren't particularly informative about what's changed.

b) can you configure the fan profile/behavior in the UEFI?

No, not that I can see.

c) ceterum censeo: is there a parallel windows installation?

Yes. It's dual boot with Win-10.

What specific model is this?

Dell Precision 5520

seth · 2024-02-01 14:24:07

Yes. It's dual boot with Win-10.

3rd link below. Mandatory.
Disable it (it's NOT the BIOS setting!) and reboot windows and linux twice for voodo reasons.

james-h-arch · 2024-02-01 14:42:29

Yes, I do:

powercfg /H off

as part of my standard build process.

james-h-arch · 2024-02-02 13:31:46

As an experiment, I tried rolling back to an earlier BIOS from 2018.

BIOS install went smoothly and no problems observed.

Fan behaviour remained the same, i.e. 2.5k rpm on a lightly-loaded system.

seth · 2024-02-02 14:13:58

One thing that optimus-manager will have done beyond the mere deactivation of acpi_call is to unload the nvidia modules.

Otherwise and assuming the system doesn't heat up and there's no point in running the fans you could try to lie to the system:

acpi_osi=! acpi_osi="Windows 2015"

https://learn.microsoft.com/en-us/windo … inacpi-osi

But as long as the fans are under firmware control and the OS isn't heating the system up, I really don't see where and what change to any software component possibly could have caused anything here.
Does this also happen on a *really* lightly loaded system, ie. when only booting the rescue.target (2nd link below)?

james-h-arch · 2024-02-06 13:54:05

I liked the tip about lying to the system!

I tried booting into console, as suggested. Left sensor watch on. Used acpi call to switch dGPU off. Nothing else.

After 10 mins or so, PCH ran a little cooler at 53C. CPU was 34C. NVME 54C.

Fan remains on at 2.5k rpm. Gaah

I'm starting to wonder if it's something mechanical. Maybe dust has built up internally and is impeding heat transfer somehow.

seth · 2024-02-06 14:59:22

Maybe dust has built up internally and is impeding heat transfer somehow.

That would affect windows as well, no?
Is the assertion that this doesn't happen w/ windows based on old data? When did you run windows last time on the system?

The heat source seems to be the nvme, they heat up under usage and 54°C isn't a problem, but it certainly heats up.
If you boot the install iso or a live distro like grml from usb and do not mount/use the nvme, how does that impact temperatures and fans?

james-h-arch · 2024-02-06 15:10:45

That would affect windows as well, no?

Yes indeed, this makes perfect sense.

Will re-measure under Win-10. Will also try the USB boot/no NVME config.

james-h-arch · 2024-02-06 16:56:15

Sharp observation Seth.

Remeasuring under Win-10 shows the same problem behaviour as Arch. Fan remains ON at 2.5k rpm. System temperatures are also similar.

Next, I booted using a USB stick with today's Arch ISO system. After installing lm_sensors I saw similar fan behaviour (2.5k rpm) and system temperatures.

NVME was powered up and visible (via lsblk), although not mounted.

dGPU was ON in this configuration.

I had a go at installing acpi-call and loading it with modprobe. Got an error when loading, which research indicated was down to a version mismatch between ISO kernel and the rolling Arch ecosystem.

Last edited by james-h-arch (2024-02-10 16:14:09)

seth · 2024-02-06 20:08:21

If you're lucky, vacuuming will do - worst case is that the heat-pipe broke…

Arch Linux

#1 2024-01-29 13:39:06

How to switch off NVidia Optimus dGPU (Maxwell)?

#2 2024-01-29 16:26:34

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#3 2024-01-29 17:35:13

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#4 2024-01-29 21:40:53

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#5 2024-01-31 13:37:19

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#6 2024-01-31 14:32:29

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#7 2024-01-31 17:00:39

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#8 2024-01-31 21:42:07

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#9 2024-02-01 12:00:33

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#10 2024-02-01 14:24:07

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#11 2024-02-01 14:42:29

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#12 2024-02-02 13:31:46

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#13 2024-02-02 14:13:58

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#14 2024-02-06 13:54:05

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#15 2024-02-06 14:59:22

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#16 2024-02-06 15:10:45

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#17 2024-02-06 16:56:15

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

#18 2024-02-06 20:08:21

Re: How to switch off NVidia Optimus dGPU (Maxwell)?

Board footer