How to run the training on GPU ?

[ZuluPro]

Hi, When I use a tensorflow-gpu environment, despite CUDA, cuDNN and some operations involve the GPU, The training is executed on CPU.

Is there a way to force it in GPU ?

Regards

[ZuluPro]

Any news ?

[rgal]

Which script are you using? train_keras_network.py? I haven’t tried this on GPU. Can you see where different operations are running with logging device placement (https://www.tensorflow.org/guide/gpu#logging_device_placement)?

[ZuluPro]

@rgal Yes I added the logging. The CPU is use from evalutate_episode, I guess in choose_action, the np.reshape starts the GPU avoidance.