Should I download CUDA5 from NVIDIA site?


Today we will install pgilinux-2013-134.tgz in a new TESLA server with CENTOS. Which version of CUDA5 if any should I download from NVidia?

Options are several redhat and fedora versions

Thanks in advance,

Alejandro Sierra

Hi Alejandro Sierra,

We ship all necessary components for building Accelerator code (OpenACC and/or CUDA Fortran). Though, during installation you need to select “yes” when prompted if you wish to install the CUDA toolkit. 13.4 does include CUDA 5 but defaults to using CUDA 4.2 when compiling. To use CUDA 5, use the flag “-ta=nvidia,cuda5.0” or “-Mcuda=cuda5.0”.

You will need to install the latest CUDA driver in order to run your binary. Since CentOS is a Red Hat clone, I’d assume you’d use the Red Hat version, but please contact NVIDIA directly if you have questions. Note that you are welcome to download CUDA 5.0. It’s not necessary for use with PGI, but object compiled by nvcc are compatible with PGI compiled accelerator code.

Hope this helps,