This isn’t a problem, it’s more of a curiosity thing.
Updated to 6.16.11 this morning, found that it was very sluggish in responding to keystrokes. Fired up btop to see what was kicking its own head in and realised it was a lack of Nvidia drivers - at least they were not at all happy.
Rebuilt, reboot and recovered normality. Fired btop back up again to make sure everything looked happy and noticed that the GPU activity was displaying a regular pattern. Assumed it was my browser polling YouTube or something, so I shut everything down - turned the Wi-Fi radios off, turned the Ethernet port off - got it down to nothing more than Wayland, ghostty and btop. Set the btop refresh rate to 1000 m and left it alone for 10 minutes or so.
Anyone have any idea what is causing the GPU to perform a steady ramp up to about 20%, then a steady ramp down to its usual 3 or 4% every 20 seconds or so.
For those not aware, we’re looking at the green lines underneath the centre line. CPU is above, GPU is below. As you can see, nothing really active on the CPU, but a fairly consistent “wake up and have a look around” activity from the GPU. Nothing running, nothing on the network, nothing using the mouse, the screen (I powered the monitor off at the wall in case the GPU was polling it to see if it had magically produced some new resolutions or something equally unlikely.
As this is the closed proprietary driver, it’s almost certainly something Nvidia is trying to do, but what? This is not a new kernel thing - it happens on previous kernels as I’ve just found out, but because I noticed it, now I can’t unsee it.
Is it really?
What is the output of modinfo -l nvidia ?
If that produces nvidia as the result then it is the closed driver,
but if it returns Dual MIT/GPL then it is the newer open driver.
I just ran btop on my desktop machine for several minutes with the same 1000ms update, and have the nvidia drivers installed. It is using the newer open driver and this is what I see. I do not see the rhythmic behavior you display, so would not attribute it to the driver for the gpu, but rather to something else you appear to have running. (I do not have ghostty installed or used). I also had my browser in use the full time.
I don’t know the differences, but the proprietary driver should only be built if you have an older gpu that is not supported by the open driver. (Assuming you have the driver installed from rpmfusion). If it is installed from anywhere else then all bets are off.
To see what is actually installed please run dnf list --installed \*nvidia\* as well as inxi -Gxx and post those results
To find out where that libva package came from you could probably run dnf4 list --installed libva-nvidia-driver.
That could possibly be interfering and I do not have that package.
If modinfo -l nvidia still returns NVIDIA then you might try rebuilding the nvidia driver with sudo dnf remove libva-nvidia-driver to remove that one then sudo akmods --rebuild --force which should remove the current module and build it anew. The akmod-nvidia package is supposed to detect which gpu is installed and build the module accordingly. The RTX 2050 gpu is supported by the open driver and the module should be properly built for that gpu.
Reboot after doing the rebuild to load the newly compiled driver.
This is an VA-API implementation that uses NVDEC as a backend. This implementation is specifically designed to be used by Firefox for accelerated decode of web content, and may not operate correctly in other applications.
Rebuild the driver so it’s now showing as Dual MIT/GPL (I’d left a macro active override in /etc/rpm/ - took a while to find that!)
Unfortunately no difference to the regular ramp-up/down of the GPU.
Just noted the update from @PG-tips (fine tea!) and that is almost certainly where I yoinked it from, although I have no recollection of doing so - I was trying to encourage Firefox to display video in a more GPU accelerated fashion a few months back as it was using a lot of CPU to play YT videos, and spinning the CPU fans up enough to be annoying. In the end I employed a hardware approach and replaced the stock cooler with a larger twin tower affair which frankly I should have done years ago.
Anyway, still got the GPU “breathing” going on, so I think I’ll boot into a lower runlevel / systemd target and see if it goes away when there’s no wayland running!
(Not sure all those preferences are strictly required.)
Then to test the settings, go to the pseudo-URL about:support#media and you should see something like this. (Exact set of supported codecs will depend on what series of GPU card you have - this is from a 3000 series)
And as expected, it’s “normal” with no wayland loaded and its even “normal” in an alternative TTY when a full system is running on the main wayland display server but not actually being displayed.
Found it. Been looking at the cause of this weirdness all day and the penny just dropped.
It’s the wallpaper. I was using the Hunyango (whatever that is) wallpaper and it changes colour every 20 seconds or so. It’s the fade in and fade out of the desktop wallpaper…
Sigh.
Also, @pg-tips where was your handy list of incantations to chant when I was losing my mind trying to get GPU acceleration in FF working better than it was! I’m using Zen now anyway, so I think I’ll have a fiddle with this lot anyway, just for fun.
