First, does Pytorch 0.2.0 still support Cudnn 5.1? I have some Torch code to run, but currently Torch only support cudnn5.1, as far as I know. SO I install cudnn 5.1 globally in my CUDA installation folder.
But when I check within Pytorch cudnn version using,
torch.backends.cudnn.version()
It shows cudnn version of 6021. But I do not have cudnn 6.0 installed. SO does Pytorch install cudnn 6.0 automatically. Or i will run into problem when I actually run some pytorch code?
Thanks for the clarification. Any plan to make Torch support Cudnn 6 or even Cudnn 7 in the future as well? I think Pytorch is very great, but I still have quite a lot torch code to run.
So if I understand this correctly, it is not necessary to download CUDNN explicitly by registering for the NVIDIA Dev program and downloading it oneself anymore?
Also last I checked CUDNN had something like per-person licensing, so how does this work?
So just to clarify:
With PyTorch version 2, there is no need for a system wide cuDNN installation? if that is the case, which cuDNN version does PyTorch ship with?