recently I see that NVIDIA has explicit support for Deep Learning frameworks, maybe all the major frameworks (Caffe, Theano,Thorch ... etc) have support for CUDA and in some cases more specific libraries are builded like cuDNN. So what is the counterpart of these in AMD/ATI ecosystem? Of course I think OpenCL is the way but the references for this is so marginals that I must do this question. Also I think if we have a counterpart of already builded in CUDA but in OpenCL we have more hardware to exploit not only the NVIDIA hardware but also the multicore systems, Intel, AMD or Parallella hardware and even NVIDIA hardware.
Thanks by your time.