How to choose nvidia cards for CUDA programming?

edit

I am using CUDA for computer vision and machine learning, like image processing, calculating HOG, training SVM, etc. I've also noticed the issues in computing from http://stackoverflow.com/questions/10765518/how-to-quantify-the-processing-tradeoffs-of-cuda-devices-for-c-kernels

So I am wondering:

1. Are there differences among the nvidia gpus when using CUDA for computer vision processing? I mean if some are good at and some are not.
2. If yes can you recommend some card models? I'm oriented on a geforce series and my budget is about $300-$500

Thanks