Nvidia .run problems to install

omlx-4
Tags: #<Tag:0x00007f0e470e3a88>

#1

I’m trying to install nvidia drivers with proprietary .run installer since OM LX 4.0 alpha1 still don’t have it.

I got the following error message:

ERROR: Unable to load the kernel module 'nvidia.ko'.  This happens most frequently when this kernel module was built against the
       wrong or improperly configured kernel sources, with a version of gcc that differs from the one used to build the target
       kernel, or if another driver, such as nouveau, is present and prevents the NVIDIA kernel module from obtaining ownership of
       the NVIDIA GPU(s), or no NVIDIA GPU installed in this system is supported by this NVIDIA Linux graphics driver release.

I’ve checked nouveau is blacklisted=not loaded, the nvidia driver I’ve downloaded is the one I need for my nvidia card. Thus, could it be that kernel devel was compiled with a version of gcc (or Clang) that differs from the one nvidia-installer is using?

Thanks


(Abucodonosor) #2

@adelson.oliveira

even when users don’t like this answer , whatever is buggy in these run files , precompiled
blobs is imposible to fix. So complain to Nvidia or use Distribution provided packages.


#3

They aren’t available for this alpha release. Also, in OM LX 3.03, I still don’t know why I can’t configure prime with nvidia-settings provided by nvidia OM packages …

But let me change my question:
Could it be that kernel devel was compiled with a version of gcc (or Clang) that differs from the gcc 8.2.0-6 available at OM LX 4.0 alpha1?


(Abucodonosor) #4

Yes but that should not matter … what matters to compile the modules is to have
same kernel installed and for that kernel same source/devel packages you compile
against.

If Nvidia really want exact version of the compiler then they are s***** , really


#5

I’m approaching a dead lock on the use of optimus and hybrid graphics laptop. Can’t use bumblebee at my optimus laptop because it has not been maintained and a new glvnd has come in, and I cannot use prime at my hybrid graphics laptop because there seems to be problems with nvidia proprietary drivers and kernel 4.20.
And nouveau is not sufficiently developed


(Abucodonosor) #6

I know , you won’t like that … but … It is know forever NVIDIA is a cra* company.
it is know forever how bad the driver support is in Linux , it is known forever is closed
source … so why do you buy such HW to start with ?

We cannot do anything about any bugs in any Nvidia drivers , blobbs etc.

The only thing can sometimes be fixed are compier error in that weird kernel driver modules
they provide. And by fix I mean ‘we guess from that code’ and ‘look kernel changes’ , compare
and ‘try a fix’ … Whatever this is right or wrong we don’t know …

That said … Bugs need go to Nvidia since we cannot do anything about.

And yes I know the Nvidia drivers in OM are not really maintained.

And nouveau is not sufficiently developed

LOL! I guess you don’t realized what nouveau is and how is developed , right ?


(Colin Close) #7

Hi,
I feel your pain. It might be worth checking whether the .run file is using
gcc for compilation. Just because it says gcc in the error message doesn’t
neccessarily mean that it was used. I have done some maintenance on the OMA
nVidia drivers previously and I can try and fix them for Lx4. I am very busy
at the moment though so I won’t be able to get to it until next week at the
earliest. You could try CC=ggx CXX=g++ ./nvidia.run or whatever the name of
the nVidia file is.
Best,
Colin


#8

Well I guess you meant CC=gcc CXX=g++ ./Nvidia…
And YES, it worked finally! Never so easy! Thanks Collin!
Now I can move for the next step: configuring prime with nvidia-settings


#9

Yes, I don’t know much about nouveaus’ development. Hope it is being done and that it can be a nice option for proprietary software as soon as possible.

At this moment, I haven’t heard about any use of nouveau in algorithms that use GPU with cuda. Recently one of my codes performed about ten times faster using libraries like CUBLAS and this seems to demand nvidia’s cuda packages.
To run my codes in a Optimus laptop with OM LX 3.0 all that I had to do was,

$ optirun mycode

optirun is provided by bumblebee, above it loads nvidia drivers and launch mycode

Thus, so far, I need nVidia, bumblebee, and in a new hybrid graphic laptop I probably will need prime.


(Abucodonosor) #10

Yes, I don’t know much about nouveaus’ development. Hope it is being done and that it can be a nice option for proprietary software as soon as possible.

It is done by ‘reverse engineering’ , eg: from dump of hardware registers , firmware , system calls
from the blobs. There is no support what so ever from Nvidia. The nouveau developers are heros … not sure you can even imagine how much work that is.


#11

And considering that every new nVidia model may require a restart from scratch … It is really amazing


#12

After having nvidia drivers installed and loaded, now I’m trying to run nvidia-settings to configure prime (maybe I’m in the wrong way, anyway…). I’m getting the following error message:

$ nvidia-settings
ERROR: Unable to load info from any available system

here is a discussion about this,

https://devtalk.nvidia.com/default/topic/1032882/linux/nvidia-settings-error-unable-to-load-info-from-any-available-system/

By now, everything looks a bit confusing, but in link above the nvidia’s guy reminds that the nvidia-installer is supposed to work with the X out.


(Tomasz Paweł Gajc) #13

We never had a support for nVidia run files.
If something is wrong please file a bug report upstream:
https://devtalk.nvidia.com/default/board/98/linux/


#14

Thanks for the tip on the nvidia’s bug report link.

I did not think OM would give support to someone else’s software. However, since there seems to be an effort to provide OM’s versions of Nvidia’s proprietary drivers in non-free repos, I thought it could be worth to report my experience on installing nvidia’s run files since bottlenecks could be anticipated …