Is new CUDA SDK always better?

My GPU is computer capability 2.1 and works with CUDA 4.2. Will I get any benefits if upgrade to CUDA 6.5?

One thing I noticed is that it seems --ptxas-options=-v option doesn’t work under 4.2 (it is set through NSight).

New SDK is supposed to be better but I doubt if Nvidia puts much effort to keep the backwards capability or fix bugs for old architecture in the new SDK. Since the new SDK is mainly for supporting new architecture maybe it will perform a little worse on old architecture. I didn’t see anything about this in CUDA release note or documents. Hope someone can give a clear answer.