Skip to content

Latest commit

 

History

History
44 lines (26 loc) · 1.52 KB

2. Installing CudaToolkit.md

File metadata and controls

44 lines (26 loc) · 1.52 KB

Prerequisites:

  1. Have/ Generate NVIDIA Developer account. https://developer.nvidia.com/
  2. Terminal: nvidia-smi
    • It might/might not return result depending if you have cuda
  3. Reboot

(Choose Your CUDA version based on your tensorflow requirements as indicated in 1. Cleaning Corrupt Dependencies. )

To insall TensorFlow 2.4.1:

A. Open: 1. Cuda toolkit 11.0 from Cuda archives : https://developer.nvidia.com/cuda-11.0-download-archive 2. Follow installation instructions for Your operating system.

For Linux | UBUNTU 18.04:

B. Ubuntu >> 18.04 >> .deb(local)

C. Follow only the first 5 steps on Terminal from the instructions.

Do not perform sudo apt-get update

D. sudo apt-get -o Dpkg::Options::="--force-overwrite" install cuda-11.0

E. sudo reboot

F. nvcc -V

>>This will return the Nvidia Cuda toolkit you intended. in case of failure, the latest version gets installed which may not be suitable for you needs. IN case of failyre, start again.

G. Check usr/local directory. Folder must have installed virsion.

H. Add file to path: Follow Installation guide on Linux post installation documentation https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#ubuntu-installation:

* echo 'export PATH=/usr/local/cuda-11.0/bin:$PATH' >> ~/.bashrc  #Dont forget to edit the cuda version which you installed
* echo 'export LD_LIBRARY_PATH=/usr/local/cuda-11.0/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
* source ~/.bashrc
* sudo ldconfig

I. sudo reboot

##Done!