TensorFlow: GPU vs CPU in CNN

CNN2.py

from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
import time


def inference(images, keep_prob):
    images_re = tf.reshape(images, [-1, 28, 28, 1])

    with tf.name_scope("Conv1"):
        W_conv1 = tf.Variable(tf.truncated_normal([5, 5, 1, 32], stddev=0.1, dtype=tf.float32), name='W_conv1')
        b_conv1 = tf.Variable(tf.constant(0.1, shape=[32]))
        h_conv1 = tf.nn.relu(tf.nn.conv2d(images_re, W_conv1, strides=[1, 1, 1, 1], padding='SAME') + b_conv1)

    with tf.name_scope("Pool1"):
        pool1_op = tf.nn.max_pool(h_conv1, ksize=[1, 2, 2, 1],
                                  strides=[1, 2, 2, 1], padding='SAME')

    with tf.name_scope("Conv2"):
        W_conv2 = tf.Variable(tf.truncated_normal([5, 5, 32, 64], stddev=0.1, dtype=tf.float32), name='W_conv2')
        b_conv2 = tf.Variable(tf.constant(0.1, shape=[64]))
        h_conv2 = tf.nn.relu(tf.nn.conv2d(pool1_op, W_conv2, strides=[1, 1, 1, 1], padding='SAME') + b_conv2)

    with tf.name_scope("Pool2"):
        pool2_op = tf.nn.max_pool(h_conv2, ksize=[1, 2, 2, 1],
                                  strides=[1, 2, 2, 1], padding='SAME')
    with tf.name_scope("Densely1"):
        W_Dens1 = tf.Variable(tf.truncated_normal([7 * 7 * 64, 1024], stddev=0.1, dtype=tf.float32), name='W_dens1')
        b_Dens1 = tf.Variable(tf.constant(0.1, shape=[1024]))
        h_pool2_flat = tf.reshape(pool2_op, [-1, 7 * 7 * 64])
        h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_Dens1) + b_Dens1)

    with tf.name_scope("Droupout"):
        h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

    with tf.name_scope("Readout"):
        W_fc2 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.1, dtype=tf.float32), name='W_dens1')
        b_fc2 = tf.Variable(tf.constant(0.1, shape=[10]))
        y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2

    return y_conv


def main():
    n_classes = 10
    n_inputs = 784
    x = tf.placeholder(tf.float32, [None, n_inputs])
    y_ = tf.placeholder(tf.float32, [None, n_classes])
    keep_prob = tf.placeholder(tf.float32)
    mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

    y_conv = inference(x, keep_prob)

    cross_entropy = tf.reduce_mean(
        tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv)
    )

    train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
    correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    start_time = time.time()
    sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))  # , device_count = {'GPU': 0}
    writer = tf.summary.FileWriter("./graphs", sess.graph)

    sess.run(tf.initialize_all_variables())
    for i in range(20000):
        batch = mnist.train.next_batch(50)
        if i % 100 == 0:
            train_accuracy = sess.run(accuracy, feed_dict={
                x: batch[0], y_: batch[1], keep_prob: 0.75})
            print("step %d, training accuracy %g" % (i, train_accuracy))

        sess.run(train_step, feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})

    test_accuracy = sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})
    print("test accuracy %g" % test_accuracy)
    end_time = time.time()
    print("time to train and output the test prediction was: %d " % (end_time - start_time))
    writer.close()


if __name__ == "__main__":
    main()


    # gpu 2 layer: 284 sec
    # cpu 2 layer: 2080 sec

    # gpu 4 layers: 953 sec
    # cpu 4 layers: 7991 sec

What about time to score the models?