# for connection and file transfer
ssh -i ~/Dropbox/research/aws_noisemodel_keypair.pem ubuntu@ec2-54-164-130-227.compute-1.amazonaws.com
rsync –progress –delete -rave “ssh -i /home/czxttkl/Dropbox/research/aws_noisemodel_keypair.pem” /home/czxttkl/workspace/mymachinelearning/Python/LoLSynergyCounter ubuntu@ec2-54-164-130-227.compute-1.amazonaws.com:~/
sudo apt-get install python-pip python-dev pip install tensorflow-gpu
download and transfer cuda toolkit, then install
sudo dpkg -i cuda-repo-ubuntu1604-8-0-local_8.0.44-1_amd64.deb sudo apt-get update sudo apt-get install cuda
download and transfer cudnn, then install:
tar xvzf cudnn-<your-version>.tgz sudo cp cuda/include/cudnn.h /usr/local/cuda/include sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64 sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
reference: https://github.com/tensorflow/tensorflow/issues/5591
Other possibly used scientific modules
sudo pip install gensim numpy scipy scikit-learn pandas seaborn sudo apt-get install python-tk
Append to ~/.bash_rc and run source ~/.bashrc
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64" export CUDA_HOME=/usr/local/cuda
reference: https://www.tensorflow.org/versions/r0.11/get_started/os_setup#optional_linux_enable_gpu_support
You can also run some p2 instance optimization specific for GPU computation:
http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/accelerated-computing-instances.html#optimize_gpu