I am attempting to install CUDA on a DELL M6400 running Fedora 10 2.6.27.21-170.2.56.fc10.x86_64 with GPU FX QUADRO 3700M
The gcc version is 4.3.2
I noticed that the website provides up to version Fedora 9, however I downloaded the 180.22 driver for this version,the SDK and toolkit and followed the installation process described.
After installing the driver and running nvidia-xconfig outside X windows I was unable to restart X windows.
I previously used the180.29 driver I got through RPMFusion.
I too am a Fedora 10 user and I got CUDA installed just fine even with the Fedora 9 references so don’t despair.
I’ve used the beta drivers on my system they can be found here download the one appropriate to your graphics card.
Firstly I would uninstall the rpmfusion rpm and that should remove the nvidia kernel modules. Making sure though that you’ve edited your /etc/inittab file and change the line:
id:5:initdefault:
to
id:3:initdefault:
so effectily going from init from 5 to 3 before you reboot. This will stop X from starting by default.
Do a reboot so that you’ve got rid of the rpmfusion’s kernel mod and then login to your account via the classic text login which will be present - iniit 3 after all sh ./nvidiafilename (you can sudo or su -c if you’ve logged in as yourself) and then it should start the nvidia install procedure. You may get a strange error at the end - this is something which has been happening recently when you install both the 32 and 64 bit opengl compoenents. But if your just 32 bit or your 64 and don’t install the 32 bit components you won’t see the error.
When that is done you can type init 5 or startx as root su or sudo and you should now have a working environment again.
If your happy then edit inittab back to 5 and go through the SDK and toolkit install. Remember you need to be root or have root permissions to install the toolkit.
One draw back with this is you will have to rebuild the kernel mod with each kernel update so keep an eye on your yum updates and don’t panic if you can’t boot into init 5 on a fresh kernel.
I hope this helps?
The rpmfusion nvidia drivers have the CUDA code hooks in them anyway, why not just use them?
I went with the nvidia “run” style drivers because I was using CUDA 2.2 Beta and needed the very latest driver so I used the proprietary one but now that it’s out officially the rpmfusion driver will eventually catch up. I might go back now that 2.2 is official because the rpmfusion one is so much handier to deal with (it auto-updates with the kernel since it has a “akmod” support )
Is there any CUDA functionality in the proprietary drivers that the rpmfusion ones lack?
