How can I check whether I use CPU or GPU in TensorFlow?

Question

I've read that

os.environ["CUDA_VISIBLE_DEVICES"] = ''

takes care that tensorflow will run on CPU and that

os.environ["CUDA_VISIBLE_DEVICES"] = '0'

takes care that tensorflow will run on GPU 0.

How can I check, which device is used?

The code

# Creates a graph. a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a') b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b') c = tf.matmul(a, b) # Creates a session with log_device_placement set to True. sess = tf.Session(config=tf.ConfigProto(log_device_placement=True)) # Runs the op. print(sess.run(c))

shows only the result

[[ 22. 28.] [ 49. 64.]]

and no used device etc.

Possible duplicate of How to tell if tensorflow is using gpu acceleration from inside python shell? — IanQ
– IanQ, Commented Nov 16, 2018 at 19:47

bivouac0 · Accepted Answer · 2018-11-17 02:42:33Z

You should be able to do this by turning on the tensorflow logging statements. There's a few ways to do this. You can do it with a bash environment variable with..

export TF_CPP_MIN_LOG_LEVEL=1

or from within your code with..

tf.logging.set_verbosity(tf.logging.INFO)

On my system I get something like...

Device mapping: /job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device /job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device /job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX TITAN X, pci bus id: 0000:65:00.0, compute capability: 5.2 MatMul: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0 a: (Const): /job:localhost/replica:0/task:0/device:GPU:0 b: (Const): /job:localhost/replica:0/task:0/device:GPU:0 [[22. 28.] [49. 64.]]

I use spyder and I noticed that the output from "print(sess.run(c))" is different to the output when I use the Python console, which is similar to the description in tensorflow.org/guide/using_gpu. Why?
I'm not familiar with spyder but it must be doing something that's suppressing the python logging. A quick search showed a few posts on this such as.. github.com/spyder-ide/spyder/issues/2572. You'll have to google and read through the posts if you want to figure fix the log suppression issue.

robertradar · Accepted Answer · 2018-11-18 14:31:44Z

I found that there are differences between Python and IPython. IPython is the used Kernel in Spyder. So I'm guessing that's the reason for those differing outputs.

Collectives™ on Stack Overflow

How can I check whether I use CPU or GPU in TensorFlow?

2 Answers 2

2 Comments

Comments

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Linked

Related