Skip to content

Latest commit

 

History

History
76 lines (58 loc) · 2.95 KB

GPU_AWS_README.md

File metadata and controls

76 lines (58 loc) · 2.95 KB
Raw

AWS GPU Installation Insructions

  • Select an Ubuntu 14.04 AMI, like ami-2d39803a.
  • Select a GPU instance, like g2.2xlarge
  • Pick a VPC and subnet
  • Set "Auto-assign Public IP" to "enable"
  • Select IAM role that can write to S3 (I made one called s3_write)
  • Add a healthy amount of storage, like 100 GB
  • Add a meaningful tag, like "Name":"ConvNet-GPU-Instance"

Run the commands below.

# Do everything as root just to make things easier
sudo su

# Update the GPU and say 'yes' to any questions
apt-get update && apt-get -y upgrade

# Not sure what this is trying to install
apt-get -y install linux-headers-$(uname -r) linux-image-extra-`uname -r`

# Get and install CUDA
wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb
dpkg -i cuda-repo-ubuntu1404_7.5-18_amd64.deb

# Upgrade/install again specifically for CUDA
apt-get update -y
apt-get install -y cuda

# Get and install CUDNN
CUDNN_FILE=cudnn-7.0-linux-x64-v4.0-prod.tgz
wget http://developer.download.nvidia.com/compute/redist/cudnn/v4/${CUDNN_FILE}
tar xvzf ${CUDNN_FILE}

cp cuda/include/cudnn.h /usr/local/cuda/include
cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

echo 'export CUDA_HOME=/usr/local/cuda
export CUDA_ROOT=/usr/local/cuda
export PATH=$PATH:$CUDA_ROOT/bin:$HOME/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_ROOT/lib64:/usr/local/cuda-7.5/extras/CUPTI/lib64
' >> ~/.bashrc

# Install Anaconda
# Python will break if you try to put it in /root, so use /mnt/bin instead
mkdir -p /mnt/bin
wget http://repo.continuum.io/archive/Anaconda3-4.0.0-Linux-x86_64.sh
bash Anaconda3-4.0.0-Linux-x86_64.sh -b -p /mnt/bin/anaconda3
echo 'export PATH="/mnt/bin/anaconda3/bin:$PATH"' >> ~/.bashrc

# Turn on bash profile to activate the PATH variables from above
exec bash

# Download and install TensorFlow binary (way easier than using source)
export TF_BINARY_URL='https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.9.0rc0-cp35-cp35m-linux_x86_64.whl'
/mnt/bin/anaconda3/bin/pip install $TF_BINARY_URL

# Install git so you can clone your repo and then run your code
apt-get install -y git    

# Get data training and validation data
wget https://s3.amazonaws.com/self-driving-car/final_processed_data_3_channels.npz
wget https://s3.amazonaws.com/self-driving-car/training.npz
wget https://s3.amazonaws.com/self-driving-car/validation.npz

# Donwload an entire S3 folder
aws s3 sync s3://self-driving-car /home/ubuntu

# Reliable command to print GPU utilization, from Nvidia
# Comes pre-installed on Ubuntu 14.04
nvidia-smi

# Optional command to write sysout and stderr to file
nohup python train_3d_conv_net.py 2>&1 | tee /home/ubuntu/log.txt &