I'm not an nvidia fanboy. OpenCL is improving and being open source and a Linux guy myself, I want it to succeed. But right now, OpenCL still has a long way to go. While more and more modern software as well as GPU are supporting OpenCL, there are still some things that you can do in CUDA that you can't in OpenCL. And should we just ditch some softwares that are a bit old? We live in the present, or at least I do, and not in the future. And right now, I'd stick with CUDA. Most people are trained in CUDA and not OpenCL. Blame schools, not me. Right now, it's relatively hard to work with OpenCL than with CUDA. I know some non-consumer products that use CUDA. My company still carry a few around.
Going forward, OpenCL would be the right path. That much I'll agree on.