Just wanted to let you know my results of porting existing CUDA application to OpenCL (http://3.14.by/en/md5).
Here are the numbers:
CUDA: 26.73 MHash/sec
OpenCL: 25.38 MHash/sec (some stuff is hardcoded, so actually it should be a tiny bit slower)
So, I’ve decided to stay with CUDA for few months, looking forward to see same performance as CUDA & working cross-platform execution.