Linux @programming.dev (⬤ᴥ⬤) @lemmy.blahaj.zone 2mo ago

how can i check for driver/gpu related problems on linux?

i've recently acquired a 4060 ti for €300 from a seller that i've been told was reliable. after installing the gpu and updating the drivers to the latest recommended ones i've noticed odd lag spikes lasting several seconds when using blender, these did not happen when i used my previous gtx 1060 ti.
How can i go about diagnosing if it's a gpu or driver issue? any benchmarks or tools i should use?

i'm on linux mint 21.3 cinnamon.

20 comments

Try launching the program from a terminal, and see if there are any GPU specific log messages. May be also have a look via journalctl.

Also nvtop and nvitop gives you a GPU monitor.
- journalctl doesn't seem to report any GPU specific errors. nvtop shows no oddities that i could find.
  
  i found that dragging the preview image on the fps showcase videos in nvidea's 4060 page makes firefox lag out with a glib-CRITICAL logged to the console. it says
  
  (firefox:4349): GLib-CRITICAL **: 12:40:07.460: g_strv_length: assertion 'str_array != NULL' failed [Parent 4349, Main Thread] WARNING: g_strv_length: assertion 'str_array != NULL' failed: 'glib warning', file /builds/worker/checkouts/gecko/toolkit/xre/nsSigHandlers.cpp:187
  
  dunno if that has to do with anything tho
Mint 21.3 might be a bit too 'stable' for your new GPU.

Linux graphics move fast. You generally won't have a good experience with an older distro and a brand new GPU.
- is there any way i can upgrade without having to reinstall from scratch?
  
  There's not really an upgrade path from Mint to a rolling release distro like Fedora or Arch. The simplest thing to do is rsync -axHAX your /home directory onto external storage and then install Fedora, Bazzite, Arch or EndeavourOS and copy data back as needed.
  
  Mint 22 would be straightforward, at least: https://linuxmint-user-guide.readthedocs.io/en/latest/upgrade-to-mint-22.html
Just had a similar issue a few days ago, where my system would freeze and spent 2 days trying to solve it. Ran a memory test, no issues. Tried NVIDIA drivers 470,535 and 550 with different combinations of the kernel including 6.x and 5.x (make sure you're using generic). Issue was still there, grabbed a copy of windows and tried it there, same issue. That narrowed it down to a hardware issue. Started tweaking the bios and found the problem was a pcie gen issue, my motherboard was automatically setting it to 5 which was causing the crash I set it to 3, no more problems. You could use the same techniques to narrow down your problem as well. (Nothing was ever reported in any logs since it was a hard freeze, which made this a huge pain in the ass to fix)
- i'll see if my motherboard has any similar options
Glmark2 is a pretty decent GPU tester for Linux, I used it just last night to diagnose memory issues on my new card. Be sure to use the -s command to set a higher resolution as the default is 800x600.
Anecdotal and likely not very helpful, when I was running a laptop with a 2060, I regularly had freezing issues launching steam, that would freeze the entire desktop on whichever I display I launched it on for anywhere from 10 seconds to a minute. The same issue occurred across multiple Linux distros running multiple Nvidia proprietary driver versions.

I built a desktop with an AMD GPU to solve the problem and it worked. I really wish I could give you something better than that.
- doubt it's the same problem i'm having because the lag spikes only affect blender, the rest of the DE works fine
  
  I guess the takeaway for me is despite making some strides in last couple years nvidia is still just kinda janky on Linux. Good luck getting to the bottom of it.
  
  Admittedly, I don't know much about the ins and outs of GPUs, but perhaps it's a GTX vs. RTX issue. They're different enough that the NVK project is targeting only RTX and up.
Maybe you can try to check if you're using legacy drivers or the new version that's party open-source.
I think Nvidia switched to those for their most recent cards not long ago.
- i'm on 550.107.02, the ubuntu-drivers command shows there are two 550 drivers one with -open after it, is that what you're talking about?
  
  Yeah, I think open should be the new one.
Sometimes the best driver comes with the kernel. It's a per model thing.
- That's not really the case with current Nvidia GPUs (yet anyway) - the Nvidia provided dkms driver is the best option outside very limited use cases

20 comments