Same failure: I guess I’m selecting too much things to be reinstalled and that leads to the “suse-prime forced comeback” (even though it is clearly removable so not necessary for those packages). Will give a new shot and come back in a few time
I can personally tell you that " Cuda + Nvidia + bumblebee + codecs “safe” way" works until Kernel 6 with recent TW.
Unfortunately I am way too busy at a new job to update the docs but I would love to see someone supporting bumblebee again.
Really? Gosh, I took my conclusions from your posts over there:
Those two were decisive for me and made me conclude that without major changes to the recipe it would not work. In the end, from the user PoV, optirun is simpler than __NV_PRIME_RENDER_OFFLOAD=1 __NV_PRIME_RENDER_OFFLOAD_PROVIDER=NVIDIA-G0 __GLX_VENDOR_LIBRARY_NAME=nvidia…
@malcolmlewis, first of all sorry for the time to answer you back this time. The idea of using the hard way worked in the end. The drawback still was that as I can’t deactivate secure boot on the bios, the installer complained about the unsigned driver and refused to install.
Unfortunately, I did not find an option for the cuda installer create a pair of keys to me, so I ended up having to accept to disable mokutil. It worked afterwards, but at the ugly price of having the boot informing twice about having an insecure system at boot time.
How can I get rid of that message? How could I have generated the pair keys in a simple way?
I am doing well, just getting beaten in a software engineer job haha. To be clear, the the method only works within Kernel 5. Does not work yet in Kernel 6 without someone investing time into it. Unfortunately I do not have the time to do this.
I guarantee you that you will clutter up your EFI partition because it’s a bit tricky and takes some experience, just be sure you provision for breaking the system before trying.
Yes, I found that instructions (I think the opensuse official help pages are even better: SDB:NVIDIA_the_hard_way:Installing_NVIDIA_with_Secure_Boot), however it seems to apply just for the nvidia driver “installator” itself, and not for the cuda “installator” (that installs the driver by default): I don’t find a way to pass such instructions in the last one.
It is not “bad”, but it is most definitely “ugly”.
So many details that make worder and hope you are right and that:
really happens in some way that we can keep the discrete cards.
Also, as I mentioned I tried too hard to install it from the repos on 15.5, so I gave up for the hard way (that worked in the first attempt provided that I disabled the mok, which is an ugly solution).
So no ideas on how to get rid of the “unsecure” messages on boot? At least not having to deal with two different “installators”, cuda and driver?
the instructions are for the nvidia driver “installator”, and not for the cuda “installator” (that calls for the “internally available” driver “installator”): the flags to tell it to “automagically” generate the keys are definitelly not available, and I’m quite certain that the ones to just pass the keys to the “installators” are also unavailable in the cuda “installator”;
I currently have an unsigned driver and mok disabled. Is it possible for me to generate those keys, import them and just then reactivate mok to lower the possibility of any issues?
Any of those being too much “unideal” and I have to learn to live with those messages.
Thanks a lot in advanc to you two (@malcolmlewis and @SJLPHI ) for all th help provided!
@johannesrs Hi just extract the run file and use command line switches, as your user run ./cuda_12.3.0_545.23.06_linux.run --help to see Or just extract, run the installer and don’t install the driver and then manually install the driver with required options.
Hi @malcolmlewis , I’m almost (at work with other issues right now) certain that I looked in the cuda installer (not “installator”, finally remembered the correct word) help and did not find anything.
I was thinking the other way around, unninstall everything, then install just the driver and later only the cuda. But isn’t that too risky of once again having a system without either (it took a long time to get to the actual state)? And also, how can I run the driver installer only from the downloaded cuda installer? Is there a way to extract it from there, so I can “simply” issue the same options as in SDB:NVIDIA_the_hard_way:Installing_NVIDIA_with_Secure_Boot for newer driver versions?
@johannesrs So, you extrat the file thus (as your user);
mkdir extracted_cuda
./cuda_12.3.0_545.23.06_linux.run --extract=$(pwd)/extracted_cuda/
cd extracted_cuda
ls NV*
NVIDIA-Linux-x86_64-545.23.06.run
So you can just use the above run file (at multi-user.target) to install manually…
You can either stick with that version (re-install after a kernel update), or just use the nvidia-installer to update (no reboot/restart X required now [new feature]) if want to move to a later version with the current cuda install.
