GPU instructions

To run on GPU:

Must have gnu set to gpu:0 in params.py and hourglass_tiny.py

Must have in hourglass_tiny.py:

import os

os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"

os.environ["CUDA_VISIBLE_DEVICES"] = "1"

Then:

$ source activate tensorflow

$ python train_launcher.py

To run from laptop at home

$ ssh <username>@hex.otago.ac.nz

$ ssh <username>@deepest

To check processes:

nvidia-smi

To kill:

Kill -9 <PID>

To copy generated data over to GPU:

scp -r hannah@OUCS1379:~/…./data new_data

To set up to work on GPU:

(Download and install Anaconda)

$ cd /tmp

$ curl -O https://repo.continuum.io/archive/Anaconda3-5.0.1-Linux-x86_64.sh

$ sha256sum Anaconda3-5.0.1-Linux-x86_64.sh

$ bash Anaconda3-5.0.1-Linux-x86_64.sh

yes yes

$ source ~/.bashrc

(Create conda env and install the GPU version of tensorflow)

$ conda create -n tensorflow pip python=3

$ source activate tensorflow

$ pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu- 1.8.0-cp36-cp36m-linux_x86_64.whl

(Set it to the right version of CUDA and CUDNN)

$ export CUDA_HOME=/usr/local/cuda-9.0_cudnn7.0

$ export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH

$ conda install