Ask Your Question

Revision history [back]

I successfully install cuda-5.5 on fedora 19 with kernel 3.12.6-200.fc19.x86_64.

I have a GEFORCE GT540M with bumblebee and bumblebee-nvidia driver (v331.20).

I have gcc-4.8.2 on my system so I had to install gcc-4.6.4. I don't want to uninstall the original compiler so I put gcc-4.6.4 in a specific folder (/opt/gcc-4.6.4) and I create an alias to modify the $PATH and $LD_LIBRARY_PATH and temporary overwrite the 4.8.2 version.

alias gcc464='export PATH=/opt/gcc-4.6.4/bin:$PATH; export LD_LIBRARY_PATH=/opt/gcc-4.6.4/lib:/opt/gcc-4.6.4/lib64:$LD_LIBRARY_PATH'

Then I downloaded the cuda_5.5.22_linux_64.run file for cuda-5.5 (on NVIDIA page) and move into Downloads directory

sudo -s
cd Downloads

I loaded the gcc464 environment and extracted the content of the main .run file

gcc464
sh cuda_5.5.22_linux_64.run -extract=$HOME/Downloads
ll
   cuda-linux64-rel-5.5.22-16488124.run
   cuda-samples-linux-5.5.22-16488124.run
   NVIDIA-Linux-x86_64-319.37.run

I have three .run files now. I don't want to install the proprietary drivers. So:

./cuda-linux64-rel-5.5.22-16488124.run

I follow the instructions on the terminal (I choose to put cuda-5.5 in /opt for example) and cuda-toolkit is installed.

I want to install also the samples to check the system. So:

./cuda-samples-linux-5.5.22-16488124.run

I follow again the instructions and choose the right cuda folder (/opt/cuda-5.5/samples and /opt/cuda-5.5 in my case).

Then I copied the original samples (contained in /opt/cuda-5.5/samples) in my Documents folder and change the write/read permissions to do some experiments.

Every time I want to use cuda I have to load the gcc464 environment and put optirun before the nvcc command. But it works for me.