keras: slow training of LSTM

Hi, I just started using keras. Awesome work! I tried to use LSTM with the following code

model = Sequential()
model.add(LSTM(4096, 512, return_sequences=True))                                                                                                                
model.add(TimeDistributedDense(512, 4096))
model.add(Activation('time_distributed_softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop')

And I compared the efficiency with char-rnn, and found that the implementation in keras is about 4 times slower than Karpathy’s (with the same batchsize). Am I doing something wrong? I’ve attached the theano profile result Thanks you!