Open ProGamerGov opened 8 years ago
I believe OP is aware of this, but just hasn't found the time to update it...
https://github.com/jcjohnson/neural-style/issues/245 https://github.com/jcjohnson/neural-style/issues/239
Yeah same problem here. I tried just running the code with 7.5 and 5.0 subbed in, but ran into some problems when trying to run the sudo cp cuda/lib64/libcudnn part. It says the directory doesn't exist.
Any clue why? It says that usr/local/cuda-7.5/lib64 is not a directory. Looking at the directories created by the tar -xzvf part, I think it might be just called cuda, and forgoing the specific number in the directory title?
It says that usr/local/cuda-7.5/lib64 is not a directory.
Are you typing in the correct command? There should be a backslash in front of usr.
sudo cp cuda/lib64/libcudnn* /usr/local/cuda-7.0/lib64/
It says that usr/local/cuda-7.5/lib64 is not a directory.
This might help you: https://devtalk.nvidia.com/default/topic/936212/cuda-setup-and-installation/tensorflow-cannot-find-cudnn-ubuntu-16-04-cuda7-5-/post/4880388/#4880388
Anybody can write a new instruction?
Currently on the cuDNN section of Nvidia's website the following versions of cuDNN are listed:
https://developer.nvidia.com/rdp/cudnn-download
The Install.md page still uses Cuda 7.0 and Cudnn v4.
In the latest newsletter/email cuDNN 5.1 RC is said to be "2.7x faster training of networks with 3x3 convolutions such as VGG".
Not sure how much better performence CUDA 7.5-8.0 along with cuDNN 5-5.1 will provide, and whether or not it will increase the size of images that Neural-Style can make.