These are good instructions here
for getting CUDA libraries and updates. Here’s a summary of what I did.
Ubuntu Server 14.04 LTS AMI (g2.2xlarge)
This is a summary of the step that I took.
You can go to cuda-downloads. Or,
if you’re doing this for ubuntu, you can just follow the
You need to make sure all the libraries get installed. This
includes CuDNN https://developer.nvidia.com/cudnn
I took a chance, and install v5. Note, v4 is recummended; however, I haven’t had
and problems with v5.
You should add the last two lines to your .bashrc
I also installed Anaconda Python 3.5 64bit
After it’s installed, update anaconda
Tensorflow for Anaconda
You have to rename the tensorflow-0.x*-cp34-cp34m*.whl file. Note that cp34 and
cp34m stands for the python version. If you’re running 3.5, you’ll need to
change this. See the steps below..
I also use Jupyter for a non-docker version as well… Here’s my
setup file. I have a /kaggle directory where my source is
kept. You’ll want to change this to whatever directory you use.
Take a look at install docker on ubuntu.
You have to create
/etc/apt/sources.list.d/docker.list and add the following
Next, you’ve got to do a lot of updates…
You’ll want to modify the docker_run_gpu.sh to allow access to the necessary
ports. Or, you can just download the fixed docker_run_gpu.sh
Once it’s modified, you’ll want to pull down the latest tensorflow gpu version.
Check out the Tensorflow docs for more info on how to grab the docker image.