TensorFlow

Installing Tensorflow

These instructions install Tensorflow into your home directory using Compute Canada's pre-built Python wheels. Custom Python wheels are stored in /cvmfs/soft.computecanada.ca/custom/python/wheelhouse/. To install Tensorflow's wheel we will use the pip command and install it into a Python virtual environments. The below instructions install for Python 3.5.2 but you can also install for Python 3.5.Y or 2.7.X by loading a different Python module.

Load modules required by Tensorflow:

[name@server ~]$ module load cuda cudnn python/3.5.2

Create a new python virtual environment:

[name@server ~]$ virtualenv tensorflow

Activate your newly created python virtual environment:

[name@server ~]$ source tensorflow/bin/activate

Install Tensorflow into your newly created virtual environment:

[name@server ~]$ pip install tensorflow

Submitting a Tensorflow job

Once you have the above setup completed you can submit a Tensorflow job as

[name@server ~]$ sbatch tensorflow-test.sh

The job submission script has the contents

File : tensorflow-test.sh

#!/bin/bash
#SBATCH --gres=gpu:1              # request GPU "generic resource"
#SBATCH --cpus-per-task=6    #Maximum of CPU cores per GPU request: 6 on Cedar, 16 on Graham.
#SBATCH --mem=32000M               # memory per node
#SBATCH --time=0-03:00            # time (DD-HH:MM)
#SBATCH --output=%N-%j.out        # %N for node name, %j for jobID

module load cuda cudnn python/3.5.2
source tensorflow/bin/activate
python ./tensorflow-test.py

while the Python script has the form,

File : tensorflow-test.py

import tensorflow as tf
node1 = tf.constant(3.0, dtype=tf.float32)
node2 = tf.constant(4.0) # also tf.float32 implicitly
print(node1, node2)
sess = tf.Session()
print(sess.run([node1, node2]))

Once the above job has completed (should take less than a minute) you should see an output file called something like cdr116-122907.out with contents similar to the following example,

File : cdr116-122907.out

2017-07-10 12:35:19.489458: I tensorflow/core/common_runtime/gpu/gpu_device.cc:940] Found device 0 with properties:
name: Tesla P100-PCIE-12GB
major: 6 minor: 0 memoryClockRate (GHz) 1.3285
pciBusID 0000:82:00.0
Total memory: 11.91GiB
Free memory: 11.63GiB
2017-07-10 12:35:19.491097: I tensorflow/core/common_runtime/gpu/gpu_device.cc:961] DMA: 0
2017-07-10 12:35:19.491156: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0:   Y
2017-07-10 12:35:19.520737: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Creating TensorFlow device (/gpu:0) -> (device: 0, name: Tesla P100-PCIE-12GB, pci bus id: 0000:82:00.0)
Tensor("Const:0", shape=(), dtype=float32) Tensor("Const_1:0", shape=(), dtype=float32)
[3.0, 4.0]

Using Cedar's large GPU nodes

TensorFlow can run on all GPU node types on Cedar and Graham. Cedar's large GPU node type, which equips 4 x P100-PCIE-16GB with GPU Direct P2P enabled between each pair, is highly recommended for large scale Deep Learning/ Machine Learning research.

Large GPU nodes on Cedar accept both whole-node jobs and single-GPU jobs. But single-GPU requests can only run up to 24 hours. The job submission script for a single-GPU job should have the contents

File : tensorflow-lgpu-single.sh

#!/bin/bash
#SBATCH --nodes=1                 # request number of whole nodes
#SBATCH --ntasks-per-node=1    
#SBATCH --cpus-per-task=6  # Total CPU cores is 24, each GPU should use up to 6 cores 
#SBATCH --gres=gpu:lgpu:1              # lgpu is required for using large GPU nodes
#SBATCH --mem=60G               # Total memory per node is around 250GB, each GPU can ask 60G
#SBATCH --time=0-03:00            # time (DD-HH:MM)
#SBATCH --output=%N-%j.out        # %N for node name, %j for jobID

module load cuda cudnn python/3.5.2
source tensorflow/bin/activate
python ./tensorflow-test.py

The job submission script for a whole node (4 GPUs) job should have the contents

File : tensorflow-lgpu-whole-node.sh

#!/bin/bash
#SBATCH --nodes=1                 # request number of whole nodes
#SBATCH --ntasks-per-node=1    
#SBATCH --cpus-per-task=24  # Total CPU cores should be 24.
#SBATCH --gres=gpu:lgpu:4              # lgpu is required for using large GPU nodes
#SBATCH --mem=250G               # memory per node
#SBATCH --time=0-03:00            # time (DD-HH:MM)
#SBATCH --output=%N-%j.out        # %N for node name, %j for jobID

module load cuda cudnn python/3.5.2
source tensorflow/bin/activate
python ./tensorflow-test.py

Packing single-GPU jobs within one SLURM job

Cedar's large GPU nodes are highly recommended to run Deep Learning models which can be accelerated by multiple GPUs. If user needs to run 4 x single GPU codes or 2 x 2-GPU codes in a node for longer than 24 hours, GNU Parallel is recommended. A simple example is given below:

cat params.input | parallel -j4 'CUDA_VISIBLE_DEVICES=$(({%} - 1)) python {} &> {#}.out'

GPU id will be calculated by slot id {%} minus 1. {#} is the job id, starting from 1.

A params.input file should includes input parameters in each line like:

code1.py
code2.py
code3.py
code4.py
...

With this method, user can run multiple codes in one submission. In this case, GNU Parallel will run a maximum of 4 jobs at a time. It will launch the next job when one job is finished. CUDA_VISIBLE_DEVICES is used to force using only 1 GPU for each code.