6.1 i386_64 PUIAS and Tesla C1060

Hello to everyone!

This is my first post on this forum so please pardon my ignorance. My knowledge of Linux is also very rudimentary
as I have been using mostly BSDs (OpenBSD mostly) and Solaris for the past twenty years.

As the title of my post indicate I am trying to use Tesla C1060 for CUDA GPU computing on 6.1 PUIAS 64 bit version
which is a Princeton clone of 6.1 RedHat. I moved to this version of Linux after having troubles to compile the
driver on Scientific Linux 5.5. SL people advised me to move to 6.1 which didn’t run well on my hardware (full of bugs including
installer and X server) so I moved to PUIAS which runs reasonable well in-spite of bugs (apparently upstream issue).

I have installed gcc as well as kernel-devel and tried run the script which came on NVidia CD with the card. The script
fails miserably and unfortunately there are no clues to why in log files.

Could anybody provide me some kind input how to install GPU driver on this or any other version of Linux? I do not mind
even switching distro as long as it is also supported by MATLAB, Maple, Mathematica, R, and Python.

I am posting this from the home so I have no access (not yet) to machine at the moment. I could provide people with dmesg, yum info
as well as NVidia log file after running installation script.

Most Kind Regards,

Hi Oko,

One thing I noticed is you are apparently using a CD that came with the card to install the driver? I would highly suggest you ditch the CD and just download the latest driver from the Nvidia downloads site. Try running that, and if it fails, send the output of the installer log.

Though I am not familiar with PUIAS, if it is a true binary compatible distro (like CentOS or SL) I would say you have a reasonable chance of it working. However, if you run into further trouble it may be worth it to switch to an officially supported distro, which would still allow you to run Matlab, Maple, Mathematica, R, and Python just fine.

I’d try it with PUIAS and the latest driver download first and see where that leads you.


Thank you so much for your response. I realized how ignorant I am after a short conversation over the weekend with my friends from Georgia State University who run successfully CUDA 4.0 on Ubuntu 10.10. I did download the newest driver from NVidia website and manage successfully to run installation scripts. However after creating xorg.conf file with NVidia-config thin X server was unable to start. I haven’t had a chance to look at that yet. One of my students might work on it. If it doesn’t work we will try Ubuntu 10.10 which works for people at Georgia State.

PUIAS should be as close to RedHat 6.1 as it gets.



Hi Oko,

Glad you got the driver to install. As for X not starting, you may need to ensure you have any conflicting modules removed, for example you may have installed xorg-x11-drv-nouveau, which you can remove. If not, you may end up with the nouveau kernel module being loaded, and it interfering with the Nvidia one you have built. If that does not work, just post the output of your /var/log/Xorg.0.log and I am sure someone can point you in the right direction.

(Of course, you don’t really need X running to use CUDA, and can just SSH into the server and use the CUDA toolkit, but if you need a graphical environment no problem. Note if you do run without X, you will need to create the device files with a script, which is outlined in the install docs.)


Thanks one more time Pete. I will try removing nouveau driver completely. I would guess that RedHat version of X window
comes with all modules pre-instelled unlike BSDs counterparts. Still the first thing on my to do list is verifying
installation and running some simple examples. Unfortunately, I have to teach today…

Thanks for all other info. We need X because of the things like MATLAB which will be installed on the machine itself.
Actually, I am not even 100% about X server until I check the documentation for NoMachine server which I selected
for GUI access. NoMachine is just compressed ssh -Y.