test_multi_gpu.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU computation time: 0:01:01.166865\n",
      "Single GPU computation time: 0:00:04.305781\n",
      "Multi GPU computation time: 0:00:02.292814\n"
     ]
    }
   ],
   "source": [
    "import sys\n",
    "sys.path.insert(0, \"/opt/DL/tensorflow/lib/python2.7/site-packages/\")\n",
    "\n",
    "from __future__ import print_function\n",
    "'''\n",
    "Basic Multi GPU computation example using TensorFlow library.\n",
    "Author: Henk Waanders\n",
    "Original author: Aymeric Damien\n",
    "Adapted from project: https://github.com/aymericdamien/TensorFlow-Examples/\n",
    "'''\n",
    "\n",
    "'''\n",
    "This tutorial requires your machine to have 2 GPUs\n",
    "\"/cpu:0\": The CPU of your machine.\n",
    "\"/gpu:0\": The first GPU of your machine\n",
    "\"/gpu:1\": The second GPU of your machine\n",
    "'''\n",
    "\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import datetime\n",
    "\n",
    "# Processing Units logs\n",
    "log_device_placement = True\n",
    "\n",
    "# Num of multiplications to perform\n",
    "n = 10\n",
    "\n",
    "'''\n",
    "Example: compute A^n + B^n on 2 GPUs\n",
    "Results on 20 POWER8 cores SMT=2 with 2 Tesla P100:\n",
    " * CPU computation time: 0:01:01.598375\n",
    " * Single GPU computation time: 0:00:04.424795\n",
    " * Multi GPU computation time: 0:00:02.402190\n",
    "'''\n",
    "# Create random large matrix\n",
    "A = np.random.rand(10000, 10000).astype('float32')\n",
    "B = np.random.rand(10000, 10000).astype('float32')\n",
    "\n",
    "# Create a graph to store results\n",
    "c1 = []\n",
    "c2 = []\n",
    "c3 = []\n",
    "\n",
    "def matpow(M, n):\n",
    "    if n < 1: #Abstract cases where n < 1\n",
    "        return M\n",
    "    else:\n",
    "        return tf.matmul(M, matpow(M, n-1))\n",
    "\n",
    "'''\n",
    "CPU computing\n",
    "'''\n",
    "with tf.device('/cpu:0'):\n",
    "    a = tf.placeholder(tf.float32, [10000, 10000])\n",
    "    b = tf.placeholder(tf.float32, [10000, 10000])\n",
    "    # Compute A^n and B^n and store results in c1\n",
    "    c3.append(matpow(a, n))\n",
    "    c3.append(matpow(b, n))\n",
    "\n",
    "with tf.device('/cpu:0'):\n",
    "  sum = tf.add_n(c3) #Addition of all elements in c1, i.e. A^n + B^n\n",
    "\n",
    "t1_3 = datetime.datetime.now()\n",
    "#with tf.Session(config=tf.ConfigProto(log_device_placement=log_device_placement)) as sess:\n",
    "sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))\n",
    "sess.run(sum, {a:A, b:B})\n",
    "t2_3 = datetime.datetime.now()\n",
    "sess.close()\n",
    "\n",
    "'''\n",
    "Single GPU computing\n",
    "'''\n",
    "with tf.device('/gpu:0'):\n",
    "    a = tf.placeholder(tf.float32, [10000, 10000])\n",
    "    b = tf.placeholder(tf.float32, [10000, 10000])\n",
    "    # Compute A^n and B^n and store results in c1\n",
    "    c1.append(matpow(a, n))\n",
    "    c1.append(matpow(b, n))\n",
    "\n",
    "with tf.device('/cpu:0'):\n",
    "  sum = tf.add_n(c1) #Addition of all elements in c1, i.e. A^n + B^n\n",
    "\n",
    "t1_1 = datetime.datetime.now()\n",
    "#with tf.Session(config=tf.ConfigProto(log_device_placement=log_device_placement)) as sess:\n",
    "sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))\n",
    "sess.run(sum, {a:A, b:B})\n",
    "t2_1 = datetime.datetime.now()\n",
    "sess.close()\n",
    "\n",
    "'''\n",
    "Multi GPU computing\n",
    "'''\n",
    "# GPU:0 computes A^n\n",
    "with tf.device('/gpu:0'):\n",
    "    # Compute A^n and store result in c2\n",
    "    a = tf.placeholder(tf.float32, [10000, 10000])\n",
    "    c2.append(matpow(a, n))\n",
    "\n",
    "# GPU:1 computes B^n\n",
    "with tf.device('/gpu:1'):\n",
    "    # Compute B^n and store result in c2\n",
    "    b = tf.placeholder(tf.float32, [10000, 10000])\n",
    "    c2.append(matpow(b, n))\n",
    "\n",
    "with tf.device('/cpu:0'):\n",
    "  sum = tf.add_n(c2) #Addition of all elements in c2, i.e. A^n + B^n\n",
    "\n",
    "t1_2 = datetime.datetime.now()\n",
    "with tf.Session(config=tf.ConfigProto(log_device_placement=log_device_placement)) as sess:\n",
    "    # Run the op.\n",
    "    sess.run(sum, {a:A, b:B})\n",
    "t2_2 = datetime.datetime.now()\n",
    "\n",
    "print(\"CPU computation time: \" + str(t2_3-t1_3))\n",
    "print(\"Single GPU computation time: \" + str(t2_1-t1_1))\n",
    "print(\"Multi GPU computation time: \" + str(t2_2-t1_2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}