# Tensorflow Installation ## Linux Tensorflow installation is still fked up as of Jan2021. There are docker containers available with right configs but I dont know how to use them. I managed to install in locally with the following steps[^1] 1. Install the right NVIDIA drivers (or hope the ones you have are good) 2. Install CUDA `sudo apt isntall nvidia-cuda-toolkit` 3. Check if CUDA is installed and what version it is. ``` $ nvcc -V nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2020 NVIDIA Corporation Built on Wed_Jul_22_19:09:09_PDT_2020 Cuda compilation tools, release 11.0, V11.0.221 Build cuda_11.0_bu.TC445_37.28845127_0 ``` 4. Check where cuda is installed ``` whereis cuda cuda: /usr/lib/cuda /usr/include/cuda.h /usr/include/cuda ``` 5. Download version of cuDNN corresponding to the cuda version from https://developer.nvidia.com/rdp/form/cudnn-download-survey. You need to login 6. Install cuDNN `tar -xvzf cudnn-11.0-linux-x64-v8.0.5.39.tgz` 7. Add the following to the .zshrc file ``` export LD\_LIBRARY\_PATH=/usr/lib/cuda/lib64:$LD\_LIBRARY\_PATH export LD\_LIBRARY\_PATH=/usr/lib/cuda/include:$LD\_LIBRARY\_PATH ``` 8. `source ~/.zshrc` ## Windows (not tested) 1. Install right NVDIA drivers (or hope the ones you have are good) 2. Use Conda to install tensorflow for gpu `conda create --name tf tensorflow-gpu` ## Windows 1. Install right NVDIA drivers (or hope the ones you have are good) 2. Install CUDA v11.2 3. Install CUDA v10.2 (yes seriously!) 4. [^1]https://towardsdatascience.com/installing-tensorflow-gpu-in-ubuntu-20-04-4ee3ca4cb75d