list_local_device tensorflow does not detect gpu - keras

Question

Is there way to check I installed GPU version of Tensorflow?
!nvidia-smi
Mon Dec 18 23:58:01 2017
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 384.90 Driver Version: 384.90 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 1070 Off | 00000000:01:00.0 On | N/A |
| N/A 53C P0 31W / N/A | 1093MiB / 8105MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 1068 G /usr/lib/xorg/Xorg 599MiB |
| 0 2925 G compiz 290MiB |
| 0 3611 G ...-token=11A9F5872A56620B72D1D5DF707CF1FC 200MiB |
| 0 5786 G /usr/bin/nvidia-settings 0MiB |
+-----------------------------------------------------------------------------+
But when I try to detect the list local devices, only CPU got detected.
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())
[name: "/cpu:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 3303842605833347443
]
Do I have to set something else to use the GPU for Keras or Tensorflow?

Answer 1

Use pip install tensorflow-gpu or conda install tensorflow-gpu for gpu version of tensorflow. If you are using keras-gpu conda install -c anaconda keras-gpu command will automatically install the tensorflow-gpu version. Before doing these any command make sure that you uninstalled the normal tensorflow .

Answer 2

You may need this shell to config your tensorflow-gpu.
You can run this, if you want to check tensorflow-gpu.
import tensorflow as tf
with tf.device('/gpu:0'):
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)
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
print sess.run(c)
The official documents: Using GPUs.

Answer 3

The simple way using tensorflow is:
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())
With Keras:
from keras import backend as K
K.tensorflow_backend._get_available_gpus()

Answer 4

I have the same problem, but everything was in this page couldn't solve my problem. I decided to update my display adapter. Follow this way:
Control Panel>Device Manager>display adapter>Right click>Update Driver
After that, you must restart your computer, But you should consider that it's not only source of your problem.

Answer 5

I ran into the subj when using a proper tensorflow-gpu docker container, and using tensorflow-gpu installed into a virtualenv within the container. Most likely this combination properly shields GPU capabilities, which otherwise are available if running python just in the container without virtualenv.

Answer 6

I think that you need to install cuda (it must show up in your nvidia-smi)
Did you check compatibility between your cuda/CUDNN version and tensorflow-gpu ?
This may help you:
https://punndeeplearningblog.com/development/tensorflow-cuda-cudnn-compatibility/