cuda_&cudnn_in_2020.md

installation help

1. get nvidia drivers from software updates ubuntu. nvidia-440
2. get local .deb file of cuda.
3. install according to instruction
4. export PATH=/usr/local/cuda-10.2/bin${PATH:+:${PATH}}
5. export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
6. Put these commands in bashrc
7. source it.
8. download cudnn tarball
9. extract and go into that directory. you will find a directroy called "cuda".
10. sudo cp cuda/include/cudnn.h /usr/local/cuda-10.2/include
11. sudo cp cuda/lib64/libcudnn* /usr/local/cuda-10.2/lib64
12. sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda-10.2/lib64/libcudnn*
13. For pytorch to run cuda you don't need to instal all this shit, just update your ubuntu drivers

Extra stuff

1. nvcc -v
2. locate cuda | grep /cuda$
3. locate cudnn | grep /cudnn$
4. https://github.com/Hardware-Alchemy/cuDNN-sample
5. https://github.com/NVIDIA/cuda-samples , make -k
6. https://github.com/NVIDIA/DALI
7. nvidia-smi
8. https://github.com/NVIDIA/
9. https://docs.nvidia.com/deeplearning/sdk/cudnn-developer-guide/index.html
10. https://github.com/tbennun/cudnn-training
11. https://github.com/cupy/cupy