How to install the nvidia driver in F12 using the run method

[plasmonics]

(Note added on 5/31/2010: As of x86_64 F13, the i686 libraries are no longer required in step 9. The x86_64 libs are sufficient. I have tried it with and without i686 and could not detect any difference.)

Before F12, it was easy to install the nvidia driver using the "run" method posted at http://www.nvidia.org. Now it is no longer straightforward. After a lot of trial and error, I got the run method to work on a Compaq Presario notebook with a 7000M nvidia GPU. The following method works for both factory and custom kernels, or if you have multiple kernels, e.g., a factory and a custom kernel. I have both the factory kernel and a custom kernel, and driver works fine on both. I am using 64-bit F15. For 32-bit, omit x86_64. All commands except 15 are run as root.

(Note: As Dies points out below, you can also use the akmod method for custom kernels. The so-called "run" method is universal, however, and usually just requires two steps (telinit and sh). You don't even have to reboot..)

(It goes without saying that if you are installing Fedora several months after it has been released, the packages on the installation media will be several months out of date with respect to the repository. Therefore, before even proceeding to step 1, you should execute "yum update" and reboot.)

1. Download the driver from http://www.nvidia.org.

2. $ setsebool -P allow_execstack on
This is now the default setting in F15, so it is ok to skip step 2. Also, skip step 2 if not using SElinux.

3. $ yum remove xorg-x11-drv-nouveau

4. $ echo "blacklist nouveau" >> /etc/modprobe.d/blacklist.conf

5. $ yum --enablerepo=rp*g remove kmod-nvidia xorg-x11-drv-nvidia-libs.{i686,x86_64}
If you had previously installed kmod-nvidia, it has to be removed, otherwise there is a conflict. Skip if you didn't.

6. $ cd /boot

7. $ mv initramfs-$(uname -r).img initramfs-$(uname -r).img.backup

8. $ dracut initramfs-$(uname -r).img $(uname -r)

9. $ yum install gcc kernel-devel-$(uname -r) mesa-libGL-devel.{i686,x86_64} mesa-libGLU

10. reboot into run level 3 as follows:
From the GRUB boot screen, press "e" (for edit). Highlight the kernel line and press "e" again. Move the cursor to the end of the kernel line and enter "3" without the quotes.
Press enter to accept the changes. Press "b" (for boot). This will boot the kernel into level 3.

11. $ sh NVIDIA-Linux-x86_64-190.42-pkg2.run
Towards the end, the nvidia script will ask if you want it to modify xorg.conf. Answer "yes".
If you have multiple kernels, including custom, all you have to do once the current setup is finished is to boot
into each of the other kernels in turn following step 10, and re-run step 11 with a "-K" switch at the end.

(Note added on 4/22/2011: In recent versions of Fedora, it is safe to skip step 12. I tried commenting these lines out in F15, F14 and it worked. It is possible that a certain version of the X server at that time could not find those modules for x86_64 and needed to be explicitly told where to find them. Such bugs have been fixed in more recent version of Fedora. Try skipping step 12.)

12. $ cd /etc/X11 && vi xorg.conf
Add the following lines to the "Files" section:
ModulePath "/usr/lib64/xorg/modules/extensions/nvidia"
ModulePath "/usr/lib64/xorg/modules/drivers"
ModulePath "/usr/lib64/xorg/modules"

13. $ cd /boot/grub && vi grub.conf
Add "rhgb vga=791 quiet" without the quotes to the kernel line.
Replace 791 with 794 if you want 1280x1024 instead of 1024x768.

F16 uses GRUB2 by default. Add the above string to the GRUB_CMDLINE_LINUX variable in /etc/default/grub. Then update the configuration using grub2-mkconfig -o /boot/grub2/grub.cfg.

14. reboot

15. $ glxinfo | grep -i opengl
It should not give errors.

(Note added on 12/15/2011: For anyone curious about a bootup message in newer kernels that the nvidia module "taints" the kernel, this is not an error, or a sign of potential problems. It is a licensing issue. Nvidia releases its "run" driver under a non-GPL license. Nouveau is GPL.)

[khaytsus]

http://fedorasolved.org/video-solutions/nvidia-yum-kmod

Done.

Minor caveat; until kwizart is around the driver is only in the testing repo, so add --enablerepo=rpmfusion\*testing\*

Installing from the .run file is just stupid.

[Dies]

http://fedorasolved.org/video-solutions/nvidia-yum-kmod

Done.

Minor caveat; until kwizart is around the driver is only in the testing repo, so add --enablerepo=rpmfusion\*testing\*

Installing from the .run file is just stupid.

I guess you missed the part where he said he was using a custom kernel, huh? Or the part where he mentioned leigh's sticky?

How stupid is that?



@ plasmonics

I'm sure someone will appreciate your post. There are good reasons to install the Nvidia driver manually. Or maybe you want to avoid installing a bunch of *-config-display's

In any case, you might want to also mention that akmods is another option.

[plasmonics]

Right. There is also the akmod method. However I am not sure it will work for a custom kernel.
The kmod method is good if you are using factory. But some people, incl. myself, sometimes
recompile for one reason or another. Others may experiment with pre-release kernels. The run
method is very flexible and works for all distros and all kernels. Unfortunately, as you can see,
it is no longer straightforward in F12.

Edit:

http://fedorasolved.org/video-solutions/nvidia-yum-kmod

Done.

Minor caveat; until kwizart is around the driver is only in the testing repo, so add --enablerepo=rpmfusion\*testing\*

Installing from the .run file is just stupid.

It is stupid that you did not read the part about custom kernel. Leigh123linux's method does not work for that case.

[khaytsus]

F12 adds an selinux command because the execstack protection has been enabled by default now... That's it, nothing else has changed since F11. You've had to blacklist nouvaeu since F11, but don't worry about that because I let the rpmfusion packages take care of that for me. There is still the step of manually rebuilding the initrd the first time regardless, only dfference is the method but the step was always there.

[Dies]

Right. There is also the akmod method. However I am not sure it will work for a custom kernel.

It should, it re-compiles the module on the fly whenever one isn't available for the running kernel. At least I thought that was the advantage, but to be honest I haven't compiled my own kernel on Fedora in a couple of releases.

F12 adds an selinux command because the execstack protection has been enabled by default now... That's it, nothing else has changed since F11. You've had to blacklist nouvaeu since F11, but don't worry about that because I let the rpmfusion packages take care of that for me. There is still the step of manually rebuilding the initrd the first time regardless, only dfference is the method but the step was always there.

Custom kernel. It will NOT work with pre-compiled modules. Are you really that dense?

[khaytsus]

It should, it re-compiles the module on the fly whenever one isn't available for the running kernel. At least I thought that was the advantage, but to be honest I haven't compiled my own kernel on Fedora in a couple of releases.



Custom kernel. It will NOT work with pre-compiled modules. Are you really that dense?

No, should I have also added "oh, I see, custom kernel, my bad" in the second post? I was responding to it no longer being "straightforward" in F12 when in reality it has one added step.

[plasmonics]

[QUOTE=Dies;1294445]It should, it re-compiles the module on the fly whenever one isn't available for the running kernel. At least I thought that was the advantage, but to be honest I haven't compiled my own kernel on Fedora in a couple of releases.

I have used kmod before, but not akmod.
I will give akmod a try on a custom kernel on my test machine and see what happens.

[GTrip]

Steps 1-11 were completed. Step 12 fails with the attached nvidia-installer.log.txt (nvidia-installer.log)

I have since done the following (as step 12 and as the final step) with success as described by leigh123linux in http://forums.fedoraforum.org/showth...ghlight=nvidia :

yum --enablerepo=rp*g install akmod-nvidia xorg-x11-drv-nvidia-libs.i686 xorg-x11-drv-nvidia-libs (WITHOUT the .x86_64)

[plasmonics]

Steps 1-11 were completed. Step 12 fails with the attached nvidia-installer.log.txt (nvidia-installer.log)

I have since done the following (as step 12 and as the final step) with success as described by leigh123linux in http://forums.fedoraforum.org/showth...ghlight=nvidia :

yum --enablerepo=rp*g install akmod-nvidia xorg-x11-drv-nvidia-libs.i686 xorg-x11-drv-nvidia-libs (WITHOUT the .x86_64)

The reason it did not work is because you used an obsolete driver. According to your log file,
you used 185.18.14. The failure had nothing to do with step 12 but with compatibility. This driver
does not work with kernel 2.6.31.5. As a test, I downloaded 185.18.14, attempted to compile it,
and was able to replicate your log file exactly. The current driver is 190.42 and works fine with step 12.

I am willing to bet that when you ran the akmod command, it pulled down 190.42 and not 185.xx. akmod is just a wrapper for the compilation that you were attempting to do manually. Had it pulled down the wrong driver, it would have failed too.

If you want to use an older driver because your GPU works better with it, you might want to try
180.60. There is a free patch for it that brings it up to speed with the newer kernels.

[jnfbatista]

Thanks for the guide, I was having some troubles installing the nvidia binaries and your guide helped a lot

[Etherus]

Hello I did use your awesome guide, and it did install the nvidia drivers without any problem the only problem I have now is somehow it does not detect my monitor anymore, it shows as "Unknown" with the nouvou driver it showed correctly my Samsung monitor and I could choose like 60 Hz etc.

Do you know what I can do to get it back?

[leigh123linux]

Right. There is also the akmod method. However I am not sure it will work for a custom kernel.
The kmod method is good if you are using factory. But some people, incl. myself, sometimes
recompile for one reason or another. Others may experiment with pre-release kernels. The run
method is very flexible and works for all distros and all kernels. Unfortunately, as you can see,
it is no longer straightforward in F12.

Edit:

It is stupid that you did not read the part about custom kernel. Leigh123linux's method does not work for that case.

Want to put money on that ?

[leigh@localhost Desktop]$ rpm -qa |grep nvidia
kmod-nvidia-2.6.31.6-153.fc12.x86_64-190.42-1.fc12.4.x86_64
xorg-x11-drv-nvidia-190.42-5.fc12.x86_64
xorg-x11-drv-nvidia-libs-190.42-5.fc12.i686
akmod-nvidia-190.42-1.fc12.4.x86_64
nvidia-settings-1.0-3.2.fc12.x86_64
nvidia-xconfig-1.0-1.fc12.x86_64
xorg-x11-drv-nvidia-libs-190.42-5.fc12.x86_64
kmod-nvidia-2.6.31.6-160.fc12.x86_64-190.42-1.fc12.4.x86_64
kmod-nvidia-2.6.32-1.fc12.x86_64-190.42-1.fc12.4.x86_64
[leigh@localhost Desktop]$ uname -r
2.6.32-1.fc12.x86_64
[leigh@localhost Desktop]$

[plasmonics]

Hello I did use your awesome guide, and it did install the nvidia drivers without any problem the only problem I have now is somehow it does not detect my monitor anymore, it shows as "Unknown" with the nouvou driver it showed correctly my Samsung monitor and I could choose like 60 Hz etc.

Do you know what I can do to get it back?

Sometimes I have had this problem myself. The driver extracts this info from the EDID of the monitor. My Sony monitor has a minor bug in the EDID. I don't worry about it. It is a string in /etc/X11/xorg.conf in the section "Monitor". You can set this string yourself.

As root, you should be able to run
$ nvidia-settings
and set the screen resolution, color depth, sync freq, overclocking, etc., and save the results in /etc/X11/xorg.conf.

---------- Post added at 08:41 AM CST ---------- Previous post was at 08:19 AM CST ----------

Want to put money on that ?

[leigh@localhost Desktop]$ rpm -qa |grep nvidia
kmod-nvidia-2.6.31.6-153.fc12.x86_64-190.42-1.fc12.4.x86_64
xorg-x11-drv-nvidia-190.42-5.fc12.x86_64
xorg-x11-drv-nvidia-libs-190.42-5.fc12.i686
akmod-nvidia-190.42-1.fc12.4.x86_64
nvidia-settings-1.0-3.2.fc12.x86_64
nvidia-xconfig-1.0-1.fc12.x86_64
xorg-x11-drv-nvidia-libs-190.42-5.fc12.x86_64
kmod-nvidia-2.6.31.6-160.fc12.x86_64-190.42-1.fc12.4.x86_64
kmod-nvidia-2.6.32-1.fc12.x86_64-190.42-1.fc12.4.x86_64
[leigh@localhost Desktop]$ uname -r
2.6.32-1.fc12.x86_64
[leigh@localhost Desktop]$

If you read my original post, you will see that I mentioned the akmod method.

My main point is that the run method is long established and works for all distros with just one command. Some distros may not have the equivalent of a pre-packaged system like kmod/akmod.

In addition, the current driver may not be suitable for some hardware. For certain GPUs, e.g., 7000M, it overworks the GPU fan, increasing the chance of a bearing failure. This is a well known problem that has been discussed in depth at the nvidia web site. So I use an older driver to which I have to apply a patch developed by some Gentoo users. I don't see how anything other than the run method would work here.

[Etherus]

Hey plasmonics, Thanks for the great post and the help with monitor I got it perfectly working now, it is actually pretty easy once you know the steps !