Trading with NEAT algorythm

I have completed a code that works with NEAT algorythm and it work well, but since the data is very large to process, it requires a lot of time before each generation will complete. SInce the NEAT algo is working without using the GPU and there is no way to be implemented , i would like to use the GPU in order to speed up the process. It will be a RL problem so it can be done with Tensorflow, Pytorch, Theano.
I already used the multiprocess so i can use all the cores of the processors but it still slow.