PDA

View Full Version : Kmod-nvidia vs Nvidia driver


Anji_fedora
26th March 2009, 06:04 PM
What is difference between these two?


These days I was using driver from NVidia. Now I have installed kmod-nvidia. I feel kmod-nvidia makes my pc slow. What is your opinion?

Hlingler
26th March 2009, 06:11 PM
Theoretically, there is no difference: the RPMs are just the same proprietary NVidia binaries, repackaged into RPM formats. However, the RPMs are of course optimized to work with Fedora, whereas the binaries are not. Plus the RPMs can be automatically updated via YUM/PackageKit/etc.

Note that installing the RPMs without first un-installing the binary driver will almost certainly cause problems, since you will effectively by double-installing the driver, with the files in different locations.

V

WIld Wolverine
26th March 2009, 06:31 PM
What is difference between these two?


These days I was using driver from NVidia. Now I have installed kmod-nvidia. I feel kmod-nvidia makes my pc slow. What is your opinion?

I see a performance decrease also, thats why I use the Nvidia proprietery driver.
The Nvidia proprietery driver offers higher performance then the kmod-nvidia driver.

Hlinger: Maybe they are theoretically the same, but in reality the Nvidia proprietery driver performs better. There also seems to be more random glitches with the kmod then with the proprietery. At least that has been my experience.

I am sure others will disagree; so just use what works for you.

JohnVV
26th March 2009, 06:40 PM
also some people ( like me) have had problems with the different libGL in the two drivers
the mesa GL in the rpm IS different than the one in the .run . Nvidia must do this inordrer for it to run on all *nix distros .
this is the problem i have
For example :
i install the .run driver , THEN i build Celestia ( from svn ) i then uninstall the .run driver and install the .rpm .
Celestia will nolonger work , it errors with a GL seg fault .
the same thing happens if i build celestia ( using the rpm driver) then install the .run driver .Celestia will not run UNTIL I REINSTALL THE MESA GL
yum reinstall mesa-libGL mesa-libGLU mesa-libGL-devel mesa-libGLU-devel

Once i reinstall mesa AFTER installing the .run everything is fine
true you can install the .run without installing the GL headers but then the headers and the gl.so's will not match .

Hlingler
26th March 2009, 06:45 PM
For most "casual" users, the convenience of automatic RPM updates probably outweighs any (possible) performance differences, which I suspect are small, if they exist at all. For users who compile graphical apps, like JohnVV and myself, the binaries pose an additional risk by over-writing system libraries. So, all-in-all, I personally prefer the RPMs, and they are the "officially" preferred method of install. YMMV, to each his own, whatever works best for you, etc., etc., etc.

As always, I applaud NVidia Corp.'s excellent overall support of Linux with quality drivers - which is really the most important issue, IMHO.

V

Anji_fedora
27th March 2009, 12:50 PM
Theoretically, there is no difference: the RPMs are just the same proprietary NVidia binaries, repackaged into RPM formats.

If both are same then why there is such a difference in performance?

Jake
27th March 2009, 02:50 PM
I prefer.run simply because I would prefer my drivers directly from Nvidia than re-packaged, as you don't know what the person who packaged it, put in there.

Plus they tend to work better I find, and on more devices strangely. In some cases the .run has worked for me, were as .rpm has not.

Plus you have not got to wait a couple of days for someone to build the driver for you.

kosmiciatakuja
27th March 2009, 03:01 PM
When comparing the RPM-packaged and straight from NVidia binary, please remember that they might be different versions! For example, the latest kmod-nvidia from rpmfusion repo is 180.29 at this moment while NVidia published both 180.41 and the newest series (in alpha mode I think) 185.13.

So, if you compare the kmod-nvidia with straight-from-nvidia binary, there might be performance differences simply because they are different versions.

However, when comparing the exact same version, the performance should be the same simply because the binary driver/module file must be identical. I think :)

Hlingler
27th March 2009, 05:48 PM
I also suspect that any performance differences may be due to different driver versions between the NVIDIA-*.run binary and the RPMs.

When making statements that such performance differences exist, it would be helpful if those assertions could be backed up with facts. Run a benchmark on both - NOT glxgears, which is not and never was intended as a benchmark tool, but rather a true benchmark: install package hardinfo and run the FBENCH (FPU RayTracing) test. Compare results. If run on the exact same driver version using both the binary and RPMs, we could know for certain whether any performance difference exists.

V

mecharuva
27th March 2009, 07:11 PM
I wouldn't mind getting my Nvidia card to just WORK.
I installed Fedora 9 and updated it all just MINUTES ago, and then I added the Livna repository.
I have an Nvidia card in my box.
I run "yum install kmod-nvidia" as root in a terminal and it says "Package not available".
So I downloaded the .run.
After getting into init 3 and running it, it bickers incessantly about missing kernel-source or kernel-devel.
I have no clue what to do here.

Hlingler
27th March 2009, 07:15 PM
su -
yum install kernel-devel kernel-headers
yum update

Re-boot.

V

mecharuva
27th March 2009, 07:28 PM
That worked.
Thanks.

JohnVV
27th March 2009, 07:50 PM
mecharuva
I installed Fedora 9 and updated it all just MINUTES ago, and then I added the Livna repository.

the kmod-nvidia is in the rpmfusion repo
they moved from livna to rpmfusion when fedora 8 went to fedora 9
that info is in the How to
" F10 & F9 Nvidia driver guides"
http://forums.fedoraforum.org/showthread.php?t=204752

Cartagena Photos - Dubai Travel Photos on Instagram - Jiamusi Instagram Photos -