NVIDIA Settings on Linux machine not acting as expected (driver version 535.129.03-Oubuntu.22.04.1)

Graphics / Linux Linux
1

Some quick history first:

I have a ROG Strix G513 (2002) with an NVIDIA GeForce RTX 3050

I have installed Mint, PopOS, Arch, and several others and have had nothing but problems with getting it to work properly under any of them.

The open source driver, nouveau, will cause Linux to crash within minutes of booting and must be disabled in the kernel parameters before installing any distro.

Once Linux is installed and the proprietary NVIDA drivers are installed I still have frequent issues:

  • programs will randomly dump core; xorg, Brave, Firefox, and logs most frequently.
  • CPU soft lock and hard lock.
  • general protection faults
  • kernel panics

the newer the kernel, the more frequently it would happen… until I finally I tried running Firefox with a dedicated GPU which resulted in weird choppy images and partially rendered tabs.
This made me open NVIDIA Settings where I selected “Prefer Maximum Performance” under PowerMizer.
This has fixed almost everything. Sort of.

HERE is the problem:

I open NVIDIA Settings and under PRIME Profiles I select NVIDIA(Performance Mode) I am prompted for my password then a little box tells me to wait while the operation completes… this takes about 25 seconds. I am then told to reboot.
On reboot the profile is now NVIDIA(Performance Mode).
When I look at PowerMizer it says “Prefer Maximum Performance” but the performance Levels are jumping all over the place and rarely sit on 3.
I have to chance the Preferred Mode to either “Adaptive” or “Auto” then back to “Prefer Maximum Performance” in order to get the performance level to stay on 3.
If I close NVIDIA SETTINGS and then open it again the Performance level may stay on 3 or it may start jumping around again.
If the performance level starts jumping around I experience crashes.

Does anyone know what causes this problem with the settings?

In startup applications:
NVIDIA X Server Settings : sh -c ‘/usr/bin/nvidia-settings --load-config-only’

Here is my nvidia-persistenced info in case it is a problem:

user@user-xtc:~$ cat /etc/group | grep nvidia
nvidia-persistenced:x:137:
user@user-xtc:~$ cat /etc/passwd | grep nvidia
nvidia-persistenced:x:127:137:NVIDIA Persistence Daemon,:/nonexistent:/usr/sbin/nologin

Also, Is there a way I can simply select the specific power level or range of power levels I want it to run at?
It would be helpful because there is obviously a power level at which my system is unstable and excluding that from the range would allow me to use auto or adaptive without crashing my system.

Thanks.

2

Odd behaviour, no idea what causes it.

Please check nvidia-smi -lgc

3

ERROR: Option -lgc is missing its value. Please run ‘nvidia-smi -h’ for help.

I am having problems finding any references to this specific error via a google search or on this site

4

Here is a screenshot of both the output you requested and just running the command without parameters.

Screenshot from 2023-11-07 14-15-00
Screenshot from 2023-11-07 14-15-00725×492 50.3 KB

5

nvidia-smi -lgc sets a clocking range, so you’ll have to give the range, of course. e.g.
nvidia-smi -lgc 300,1000

6

Thanks… I will wait until I have a moment to think before throwing stuff at the wall next time.

7

I have taken some time and been playing with the command to grab some information about my graphics card.
I have found something that looks wrong to me.I was hoping you might confirm it or be able to explain why it is not wrong.

I used “nvidia-smi -q -d SUPPORTED_CLOCKS” and looking at the values here is what I get:

Mem Clock Speed | Min Graph Clock | Max Graph Clock

405                405                 420
810                405                 2100
5501               405                 2100
6001               405                 2100

But in NVIDIA PowerMizer the Min Graphics Clock for each Performance Level is set at only 210 MHz.

Would this not be a problem?
If the settings being used by the driver allow the minimum graphics clock speed to drop below the values displayed
by “nvidia-smi -q -d SUPPORTED_CLOCKS” would that explain my systems instability?

Thanks.

8

SUPPORTED_CLOCKS is something completely different, it’s for setting APPLICATION_CLOCKS with nvidia-smi for cuda. Ancient setting, rarely used.