Test GPU functionality - Githubissues

NLeSC / mcfly

A deep learning tool for time series classification and regression

Apache License 2.0

362 stars 82 forks source link

Closed vincentvanhees closed 6 years ago

vincentvanhees commented 8 years ago

[x] log on to DAS5 and clone mcfly
[x] install requirements: "pip3 install -r requirements.txt --user" was the command I used
[x] open a notebook and run the simple forecasting example using Keras/Theano
[ ] Activate GPU utilization
[ ] run mcfly with pamap2

dafnevk commented 7 years ago

In order to request a node with GPU on the DAS-5:

srun -N 1 -C TitanX --gres=gpu:1  --time=00:15:00 --pty bash
module load cuda80/toolkit

Then install tensorflow-gpu:

conda create -n mcfly
source activate mcfly
conda install pip numpy tensorflow  tensorflow-gpu
pip install mcfly

Make sure tensorflow is set as default backend in ~/.keras/keras.json. Now mcfly/Keras will should with GPU automatically.

cwmeijer commented 6 years ago

We used GPUs for the workshop last year so this was thoroughly tested then.