Noticeable flickering around recently changed screen areas on multiple laptops with NVIDIA PRIME

Graphics / Linux Linux
1

My workplace is experiencing very noticeable graphical flickering on multiple NVIDIA-powered laptops using NVIDIA PRIME. When doing fairly typical desktop work (scrolling through websites in Google Chrome, typing commands into Konsole, etc.), the area around whatever part of the screen is changing seems to “flicker” a bit, as if that part of the screen briefly reverted back to the contents of a prior frame before jumping back to the updated contents.

System details:

  • This is happening on at least two different NVIDIA GPUs, both the GeForce RTX 4070 laptop GPU, and the GeForce RTX 5070 Ti laptop GPU.
  • Both systems are displaying the problem on the laptop’s internal display, when the display runs at 240Hz.
  • Both systems use an Intel CPU (the 4070 system uses an i9-14900HX, and is the one I’m typing on at the moment, the 5070 Ti system is using a Core Ultra 9 275HX).
  • Both systems use NVIDIA PRIME, and the laptop internal display shows up as a PRIME display.
  • Both systems are running Ubuntu 24.04.2 LTS with the KDE Plasma desktop. The kernel in use is Ubuntu’s 6.11.0-1021-oem kernel.
  • The X server is used for graphics display, we are NOT using Wayland.
  • We’ve observed the issue with multiple NVIDIA 570 and 575 driver versions, the driver I’m using (and experiencing the problem with) at the moment is 575.57.08 with open-source kernel modules. We’ve also seen similar flickering with 550 and and 565 drivers in the past, though it’s unclear if that’s the same issue as this or not.
  • Here’s the output of the nvidia-bug-report.sh script from the 4070 system: nvidia-bug-report.log.gz (1.7 MB)

To reproduce:

  • Download the latest available Linux NVIDIA drivers from Download The Latest Official NVIDIA Drivers and install them.
  • Reboot the system to ensure the newly installed kernel modules are in use
  • Open a terminal window (I’m using Konsole), and run nvidia-smi. Ensure the expected driver version is in use (for me this is 575.57.08).
  • Open Google Chrome.
  • Go to the NVIDIA developer forums, and click “New Topic” in some forum category.
  • Start typing into the main text area on the left. You will notice the area around your text cursor flicker. If you look at the rendered Markdown on the right, you’ll notice the area where new text is appearing as you type will also flicker.

Other applications this happens in include Konsole (where it is very severe in some instances), Kate (where it usually isn’t that severe), and LibreOffice Writer. I have not been able to make flickering occur in Firefox or IntelliJ IDEA Community Edition so far. I’ve also noticed it in Google Chrome when scrolling a page (the entire screen will appear to “jump” backwards for a split second before jumping back to where it should be).

Things we’ve tried that didn’t resolve or improve the issue:

  • Enabling Vulkan in chrome://flags. This seemed to have no effect or only marginally improved things.
  • Disabling Resizable BAR via NVreg_EnableResizableBar=0. This seemed to make the problem worse.
  • Disabling GSP firmware via NVreg_EnableGpuFirmware=0 when using proprietary drivers. This also seemed to make the problem worse.
  • Preventing the GPU from powering down via NVreg_DynamicPowerManagement=0. This appeared to do nothing.
  • Setting xe.enable_psr=0 in the hopes that maybe fiddling with the Intel graphics side of things would help (since the Intel framebuffer is involved when using PRIME). This seemed to do nothing.

Things we’ve tried that did appear to resolve the issue at least partially:

  • Setting the preferred mode in PowerMizer settings to “Prefer Maximum Performance”. (NVIDIA X Server Settings → GPU 0 → PowerMizer → PowerMizer Settings → Preferred Mode → Prefer Maximum Performance). This immediately and completely resolves the issue, but also greatly increases GPU power draw according to nvidia-smi, which makes it unsuitable for our use case.
  • Reducing the display refresh rate to 60Hz. This appears to completely resolve the problem, but obviously at the cost of substantially reduced display smoothness.
  • Using the MUX switch in the laptop to switch to the NVIDIA GPU only. This cuts NVIDIA PRIME out of the picture, and makes the flickering stop on the 5070 Ti system. I have not yet tested it on the 4070 system.
  • Blacklisting NVIDIA graphics drivers entirely and switching to Intel graphics alone. This isn’t a good solution for obvious reasons, but I felt it was a valuable data point to add.
  • Switching to Wayland. This isn’t an option for us in production yet, but in a Wayland session, all flickering seems to be gone across all apps. (I have verified that the apps are indeed using NVIDIA graphics by using nvidia-smi.)

Worthy of note, on the 4070 system, NVIDIA’s 535 drivers are fully capable of running at 240Hz without flickering, even when NVIDIA PRIME is involved. The 535 drivers don’t work on 5000-series graphics though, so staying on them isn’t possible on all hardware we’re using. I have not yet bisected to find the exact driver version that introduces the flickering.

2

I have the 570.153 drivers installed and have verified the tearing occurs on all prime X11 screens unless they are set to their lowest refresh rate. This includes 2 external displays (30 or 60 Hz) and a laptop panel (60 or 144 or 240 Hz). This is with an RTX 5070 Ti Mobile GPU.

Screens where the refresh rate set to an integer multiple of the base rate seem to slowly cycle between no tearing, then increasing to bad tearing, and then back to no tearing, as in a sine wave. This cycle seems to last about 60s. However, a non-integer refresh rate like 144 Hz seems to cycle much faster.

All this implies this regression may be because the drivers may be using the “base” refresh rate of the display instead of using the the actual refresh rate of the prime screen. If so, hopefully this might be an easy to find and fix.

Of course, this is just an educated guess, but I hope that it is useful.

2 Likes
3

I was able to verify that the NVIDIA 535 open-kernel drivers do not have flickering problems with the 4070 laptop GPU. (Yes, I did pass NVreg_OpenRmEnableUnsupportedGpus=1 to make those drivers work, and yes, I am sure the NVIDIA GPU is in use.) I tried diffing the kernel module code between the 535 and 570 open-source drivers, but didn’t spot anything in particular that looked wrong (though granted I didn’t read the whole diff, as it’s over 80,000 lines long). I don’t know if this helps pinpoint the issue any better, but maybe this is helpful to know.

1 Like