Created
September 16, 2016 23:08
-
-
Save yaroslavvb/b73ff35424dd7ab762234620cf583aac to your computer and use it in GitHub Desktop.
Example of restricting part of graph to run on single core
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# try running cpu intensive test on two devices | |
import tensorflow as tf | |
import time | |
def matmul_op(): | |
"""Multiply two matrices together""" | |
n = 2000 | |
a = tf.ones((n, n), dtype=tf.float32) | |
return tf.matmul(a, a)/n | |
slow_op = matmul_op | |
with tf.device("/cpu:0"): | |
one = slow_op() | |
with tf.device("/cpu:1"): | |
another_one = slow_op() | |
config = tf.ConfigProto(device_count={"CPU": 2}, | |
inter_op_parallelism_threads=2, | |
intra_op_parallelism_threads=1) | |
config.graph_options.optimizer_options.opt_level = -1 | |
sess = tf.Session(config=config) | |
two = one+another_one | |
# pre-warm the kernels | |
sess.run(one) | |
start = time.time() | |
sess.run(one) | |
elapsed_time = time.time() - start | |
print("Single op: %2.4f sec"%(elapsed_time)) | |
start = time.time() | |
sess.run(two) | |
elapsed_time2 = time.time()-start | |
print("Two ops in parallel: %.2f sec (%.2f times slower)"%(elapsed_time2, | |
elapsed_time2/elapsed_time)) |
I multi session mean multi threading ?
is each one session run in different thread ?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hi yaroslavvb,
why do we need to pre-warm the kernel? I think it is for more accurate result. If it is so, what is the reason behind it?