After installing the latest fedora (KDE) I’ve installed the nvidia drivers according to this https://rpmfusion.org/Howto/NVIDIA. Sadly since that the system monitor constantly displays 0.0% usage for my GPU even while gaming. I have an RTX 3080.
The driver seems to work well as running “modinfo -F version nvidia” displays “580.119.02” also I’ve noticed huge FPS boost and visual quality in the game I’ve tried.
I found this post describing the same issue but the solution there was to use something other than system monitor which I don’t want.
With the latest KDE Plasma and a 3060Ti I don’t have this issue (i.e. usage displays fine). So there must be some particular set of conditions that causes it.
System monitor doesn’t show any errors neither on the GUI nor in the logs.
sudo journalctl -b _PID=23006
Jan 18 17:15:29 fedora plasma-systemmonitor[23006]: QWaylandGLContext: Failed to create the decorations EGLContext. Decorations will not be drawn.
Are you running a multi-gpu system (i.e. do you have integrated graphics as well)?
If the second point applies, it might be worth trying to disable the integrated GPU to see if it causes the system monitor to begin reporting the dedicated GPU instead. I will acknowledge that I do not use KDE, so I am unfamiliar with the finer details for the system monitor there.
I know you mentioned not wanting another application, but I quite like this tool, Resources, as it will report multiple GPUs if they are present. Helpful for showing what resources are actually in use at any given time.
It seems it indeed has integrated graphics . Now I have a strong suspicion that is the problem (also considering I believe it worked before installing the driver). I’ll try to verify this later as I can’t restart my system at the moment. If that’s the case can this help solving the bug ticket? If so I’ll also happily add it to the ticket.
Thank you for your help so far, really appreciate it.
Just a hypothesis, but since these are the proprietary Nvidia drivers, could you try installing the cuda package that includes the nvidia-smi tool? I wonder if the system monitor needs that to poll for usage info.
Still I don’t understand I thought cuda was mainly needed for AI? Why did this help and why proprietary Nvidia drivers weren’t enough? Cuda is also proprietary no?
Still I don’t understand I thought cuda was mainly needed for AI? Why did this help and why proprietary Nvidia drivers weren’t enough? Cuda is also proprietary no?
To elaborate on this a little bit, the Cuda package includes the nvidia-smi tool that, when run, dumps out a lot of diagnostic info about the Nvidia hardware running in the machine. Why it’s not included in the primary driver package I’m not 100% sure, perhaps because it’s assumed to be optional and not required for normal use.
Nvidia drivers are a moving target for lots of Linux software, so rather than trying to read sensors themselves they just piggy-back off of nvidia-smi output to make it easier.
Not quite.
Cuda makes us of the graphics processing cores in the GPU to perform lots of different calculations. The ability of the nvidia GPU cores to perform calculations much faster than the CPU makes them ideal for AI, but also for many other uses.
I agree that cuda should be included with the basic nvidia driver installation but that is another discussion.
My recommendation has always been to install both akmod-nvidia and xorg-x11-drv-nvidia-cuda at the same time to avoid the type of issues you were seeing.
. Now I have a strong suspicion that is the problem (also considering I believe it worked before installing the driver). I’ll try to verify this later as I can’t restart my system at the moment. If that’s the case can this help solving the bug ticket? If so I’ll also happily add it to the ticket.
