the openCL branch of caffe has had a lot of work recently and is usable now. Still a bit slower than with CUDA/CuDNN but works https://github.com/BVLC/caffe/tr

the openCL branch of caffe has had a lot of work recently and is usable now. Still a bit slower than with CUDA/CuDNN but works https://github.com/BVLC/caffe/tree/opencl https://github.com/BVLC/caffe/tree/opencl