Is cuDNN necessary?

I train detectnet on where 11.4
I don’t have cuDNN installed.
Will installing cuDNN give me additional hardware acceleration?

It would depend on how you are training it.

If you are using a framework like pytorch or TF, and you intend to train on the GPU, then cuDNN is necessary. if you are doing it using DIGITS, then nothing else is necessary beyond the DIGITS install.