Any recent driver version will do. Only OpenCL 1.0 is supported, and it doesn’t seem like NVIDIA will offer any further support… The OpenCL standalone code samples are here: https://developer.nvidia.com/opencl
As I pointed in another thread today, NVIDIA has not updated these samples to work out of the box without path changes/modifying the corresponding makefiles for CUDA toolkit versions >= 5.5. The last version of CUDA that contains the OpenCL samples with the SDK is 4.2, so I’d recommend that as a starting point so that you don’t have to waste time trying to ‘fix’ the samples.
As far as environment… don’t code much in Linux, so I can’t expand on that.