I recently acquired a GTX 1050 Ti from a good friend of mine, and installed it in a lower slot than my RX 6900XT. I primarily wish to use this 1050 Ti as a video encoder for streaming, recording, and rendering video, and so after installing I ran neofetch and saw that it was physically installed correctly and is being read by the system correctly. I’d hoped this was all I needed to do, but my naivety got the better of me. I needed to install drivers for it (cuz duh), so I installed the nvidia and nvidia-settings packages. After rebooting, I was able to record and render video using the 1050 Ti, but was unable to use the RX 6900XT for gaming.
Looking at Info Center > Graphics > Vulkan, I saw the problem.
After doing some research, I found the Nvidia drivers set GPU id to FASTEST_FIRST by default, instead of going by PCI_BUS_ID, like one would expect. FASTEST_FIRST guesses which GPU installed is faster by using a simple heuristic, according to Nvidia documentation. This is an environment variable one can change, but I can’t figure out how. I’ve read a ton recently on the subject, but I just can’t quite figure it out. Can I get some help?
I’m merely guessing here, but maybe you can set that variable in file /etc/environment?
Command man environment may help more.
Also, command man systemd.environment-generator might help.
I tried setting the variable in /etc/environment but that didn’t do anything, sadly. Then, i checked the man files for environment and systemd.environment-generator and experimented for a while with their instructions, but that didn’t help either. I saw a post on Nvidia’s developer forums that said to do this: export CUDA_DEVICE_ORDER=PCI_BUS_ID and it stuck kinda?
running echo $CUDA_DEVICE_ORDER returned PCI_BUS_ID which is what I need, but nothing’s happening. Info Center still shows the 1050 Ti as GPU 0, and games only run on that card. My 6900XT is doing something tho, flicking between 0 and 4% usage.
At this point, I’m not sure what else I could try. Different drivers, perhaps? I really don’t know what to do.
I’ve spent the last several days on this, I can’t remember exactly what I did and when, but here’s what I did:
first, i reached out to a friend of mine who is a huge linux nerd, and followed the suggestion she gave
it’s worth a try setting CUDA_DEVICE_ORDER=PCI_BUS_ID in ~/.profile if u havent already, though i doubt that it’s controlled per user
Nothing happened with this.
I tried export CUDA_DEVICE_ORDER=PCI_BUS_ID, and nothing happened there either, but it did set something. Typing echo $CUDA_DEVICE_ORDER returned PCI_BUS_ID.
Reading up on manuel’s suggestions, i modified /usr/lib/systemd/user-preset/90-systemd.preset but it seems to have reverted itself on reboot. Other similar files were unreadable. Not sure if that’s a format thing, or a nano thing.
I read up on this page and watched this video. The cuda developer documentation helped at least to understand what my system is doing, but not really to change it? I don’t know how to make use of the information there. The video kinda helped, but there wasn’t much I could do. Followed some of the things he talked about, but nothing happened.
I looked at my ~/.bashrc profile and wasn’t sure if I could do anything there to help.
I tried __NV_PRIME_RENDER_OFFLOAD=0 and many variants of that, rebooting with each, and still nothing changed.
I realize this is a fairly advanced and niche thing I’m trying to do, and I may not be ready for it, but I want to try to make it work to the best of my ability.
Oh, and sorry about earlier in the thread, i misread your post intially
