为什么我在multi_gpu_model的Keras中的训练速度比单gpu差？

Question

My Keras version is 2.0.9, and using tensorflow backend.

I tried to implement multi_gpu_model in keras. However, training with 4 gpus was even worse than 1 gpu in practice. I got 25sec for 1 gpu, and 50sec for 4 gpus. Could you give me the reason why this happens?

/blog for multi_gpu_model

https://www.pyimagesearch.com/2017/10/30/how-to-multi-gpu-training-with-keras-python-and-deep-learning/

I used this commend for 1 gpu

CUDA_VISIBLE_DEVICES=0 python gpu_test.py

and for 4 gpus,

python gpu_test.py

-Here is source code for training.

from keras.datasets import mnist
from keras.layers import Input, Dense, merge
from keras.layers.core import Lambda
from keras.models import Model
from keras.utils import to_categorical
from keras.utils.training_utils import multi_gpu_model
import time

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(60000, 784)
x_test = x_test.reshape(10000, 784)
y_train = to_categorical(y_train)
y_test = to_categorical(y_test)

inputs = Input(shape=(784,))

x = Dense(4096, activation='relu')(inputs)
x = Dense(2048, activation='relu')(x)
x = Dense(512, activation='relu')(x)
x = Dense(64, activation='relu')(x)
predictions = Dense(10, activation='softmax')(x)
model = Model(inputs=inputs, outputs=predictions)
'''
m_model = multi_gpu_model(model, 4)
m_model.compile(optimizer='rmsprop',
              loss='categorical_crossentropy',
              metrics=['accuracy'])
m_model.summary()
a=time.time()
m_model.fit(x_train, y_train, batch_size=128, epochs=5)
print time.time() - a
a=time.time()
m_model.predict(x=x_test, batch_size=128)
print time.time() - a
'''
model.compile(optimizer='rmsprop',
          loss='categorical_crossentropy',
          metrics=['accuracy'])
model.summary()
a=time.time()
model.fit(x_train, y_train, batch_size=128, epochs=5)
print time.time() - a
a=time.time()
model.predict(x=x_test, batch_size=128)
print time.time() - a

And this is gpu state with running 4 gpus.

Answer 1

I can give you what I think is the answer, but I don't have it fully working myself. I was tipped off onto this by a bug report , but in the source code for multi_gpu_model it says:

    # Instantiate the base model (or "template" model).
    # We recommend doing this with under a CPU device scope,
    # so that the model's weights are hosted on CPU memory.
    # Otherwise they may end up hosted on a GPU, which would
    # complicate weight sharing.
    with tf.device('/cpu:0'):
        model = Xception(weights=None,
                         input_shape=(height, width, 3),
                         classes=num_classes)

I think this is the problem. I'm still working on making it work myself, though.