Optimizing advantage since cuDNN 5.1?

[Timo-hab]

Since the class 'optimize' has been written, a new version of cuDNN has come out (cuDNN v5 and cuDNN v5.1).
I think cuDNN does the similar trick, that it tries to optimize the memory behavier of the deep neural networks. So, my question is, is there still an advantage compared to cuDNN v.5.1 ?
Thank you in advance!