I’m setting up a development environment which uses a linux remote gpu cluster at our university and a linux local machine with no nvidia gpu. This is the sync remote builds functionality as shown here,
The cuda toolkit 7.5 installation guide for linux only covers the cross platform environment (cross compile on host and run on target) setup. Since I have no need to install a cross architecture I didn’t.
My local system is Linux Mint 17.3 which maps to Ubuntu 14.04 and the remote is also Ubuntu 14.04 with cuda toolkit 7.5 installed.
The information for installing from a local deb worked fine, but of course with no nvidia gpu I got an error report to that effect when installing cuda. I ignored the error, set up the paths, and installed the samples locally (for reference).
Currently I’m working on getting the samples to compile and run on the remote with nsight.
I think it would be a great idea to have a use case for the sync remote builds setup that covers beginning to end, with compiling and running sample code on the remote.