Keras not detecting GPU, but tensorflow is

Question

I am trying to get my keras back-end to work, but it keeps telling that there is no GPU, but my tensorflow say otherwise.

tensorflow

from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())

Ouput

[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 7600201089948511065
, name: "/device:XLA_CPU:0"
device_type: "XLA_CPU"
memory_limit: 17179869184
locality {
}
incarnation: 9640434864482537344
physical_device_desc: "device: XLA_CPU device"
, name: "/device:XLA_GPU:0"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 15836333527683385212
physical_device_desc: "device: XLA_GPU device"
]

keras

from keras import backend as K
K.tensorflow_backend._get_available_gpus()

Output

[]

Tensorflow-gpu works, but keras can't see for some reason

Cuda is installed

tensorflow-gpu = 1.14.0
keras = 2.0.8
python = 3.6

Are you using any conda/virtualenv environment? Can you check if there are any global installations? — postmalloc
– postmalloc, Commented Sep 10, 2019 at 7:55
I'm having the same problem. Have you tried uninstalling keras and installing keras-gpu? Are you using tf.keras to build and train models? — curator23
– curator23, Commented Oct 9, 2019 at 18:32

Akshay Sehgal · Accepted Answer · 2019-09-10 09:13:19Z

2

Few things that you need to confirm before this,

Which GPU are you using? Only NVIDIA works. Do you have tensorflow-gpu installed (keras uses this as its backend)? Do you have CUDA installed?

Since tensorflow is detecting GPU, then its either an issue with on of the above points, or you may try upgrading the various packages.

Lastly, if nothing else works I would recommend using the high level API in tensorflow2 which is actually the same as keras.

Hope this helps.

answered Sep 10, 2019 at 9:13

Akshay Sehgal

19.4k3 gold badges26 silver badges57 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Currently_Lost Over a year ago

I am using nvidia, tensorflow-gpu is installed

Akshay Sehgal Over a year ago

what about CUDA?

Suraj Rao · Accepted Answer · 2019-10-10 08:47:50Z

0

Add ppa to ubuntu repos, they have the newest driver. Risky, but worth it!

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt update

Try to install newest cuda driver

sudo ubuntu-drivers autoinstall

If this fails (like it did in my case), do sudo apt install for all dependencies it complains about.

In my case:

sudo apt install libnvidia-gl-435 nvidia-dkms-435 nvidia-kernel-source-435

Tried again

sudo ubuntu-drivers autoinstall

reboot computer

Create a new conda env

conda create -n tf python=3.7

source activate tf


conda install -c anaconda keras-gpu tensorflow-gpu jupyter pytables pandas seaborn scikit-learn gensim -y

edited Oct 10, 2019 at 8:47

Suraj Rao

29.7k11 gold badges96 silver badges104 bronze badges

answered Oct 10, 2019 at 8:44

Currently_Lost

1001 silver badge9 bronze badges

Collectives™ on Stack Overflow

Keras not detecting GPU, but tensorflow is

2 Answers 2

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related