0

Please help! Using conda install tensorflow-gpu, it help me to install cuda and cudnn automatically, which is really convienience. However, the problem is that conda can only install the 7.3.1 cudnn, which is not enough for me to run the CNN in tensorflow1.13.0. Actually, I can install the newest cudnn outside easily, but it won't influence the conda environment. So how can I solve it? Thanks in advance

talonmies
  • 70,661
  • 34
  • 192
  • 269
LinTIna
  • 481
  • 1
  • 6
  • 14

1 Answers1

1

Try building a new conda env after you install cuda and cudnn (mind to find the right versions). See these steps. Also, try pip install tensorflow-gpu under the conda env (source activate env_name), too. Hope it helps!

Everley Tseng
  • 94
  • 3
  • Thank you for your response. There is one problem: I have already installed cuda9 in my computer, which is added to my path. So actually I want build cuda 10 within a conda env without influence the outside. ```conda install cudatoolkit``` help me to install cuda only for this conda env, and I need to replace the old version cudnn, but I don't know how to do that since I don't know how conda install the cuda and where is it – LinTIna Mar 30 '19 at 10:31
  • If you want different cuda versions coexisting in one device, Anaconda isn't enought. You should use Docker to manage the environment. – Everley Tseng Mar 31 '19 at 05:42
  • @everley-tseng I dont think installing CUDNN 7.5 directly on the OS will influence how conda installs the new environment. As far as I can tell, there is no option for installing CUDNN 7.5 in anaconda right now. Version 7.3.1 installs if you do 'conda install tensorflow-gpu' which is given in your tutorial link. – Adeel Hashmi Apr 27 '19 at 03:17