sess.run() multiple ops vs multiple sess.run()

Question

In tensorflow, if we want to run multiple operations, are there any differences (eg in speed or results) if we pass the ops in a list to a single sess.run() , versus we use multiple sess.run() but only pass a single ops to each one?

Example:

#!/usr/env/bin python
import tensorflow as tf

if __name__ == '__main__':
    v = tf.Variable(0)
    c = tf.constant(3)
    add = tf.add(v, c)
    update = tf.assign(v, add)
    mul = tf.multiply(add, update)
    init = tf.compat.v1.global_variables_initializer()

    # method 1
    with tf.Session() as sess:
        sess.run(init)
        print(sess.run(add))
        print(sess.run(mul))

    # method 2
    with tf.Session() as sess:
        print(sess.run([init, add, mul]))

Edit: Initially I was also wondering if there are cases where multiple Session will cause different results, eg dependencies in computational graph causes execution order to be different etc. But I didn't make this clear in my question and only gives a rather dummy example instead...

Answer 1

Each time you run the init , all the weights are initialised randomly, So no comparison can be made this way. If you really want to compare, I would run the following code:

#!/usr/env/bin python
import tensorflow as tf

if __name__ == '__main__':
    v = tf.Variable(0)
    c = tf.constant(3)
    add = tf.add(v, c)
    update = tf.assign(v, add)
    mul = tf.multiply(add, update)
    init = tf.compat.v1.global_variables_initializer()

    # method 1
    with tf.Session() as sess:
        sess.run(init)
        print(sess.run(add))
        print(sess.run(mul))
        print(sess.run([add, mul]))

Answer 2

There is no difference whatsoever in results:

import numpy as np

tf.reset_default_graph()
v = tf.Variable(0)
c = tf.constant(3)
add = tf.add(v, c)
update = tf.assign(v, add)
mul = tf.multiply(add, update)
init = tf.compat.v1.global_variables_initializer()

def single():
    with tf.Session() as sess:
        sess.run(init)
        fetch_add = sess.run(add)
        fetch_mul = sess.run(mul)
    return fetch_add, fetch_mul

def multiple():
    with tf.Session() as sess:
        sess.run(init)
        fetch_add, fetch_mul = sess.run([add, mul])
    return fetch_add, fetch_mul

add_single, multiply_single = single()
add_multiple, multiply_multiple = multiple()

np.testing.assert_array_equal(add_single, add_multiple)
np.testing.assert_array_equal(multiply_single, multiply_multiple)

When it comes to speed, here are some metrics:

%timeit single()
# 7.44 ms ± 143 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

%timeit multiple()
# 5.56 ms ± 121 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

So it is definitely faster if you run multiple nodes together.