All of the comparable tools are practically locked into nVidia gpus/CUDA. TensorFlow is rapidly reaching performance parity on that hardware [1] and can now be

All of the comparable tools are practically locked into nVidia gpus/CUDA. TensorFlow is rapidly reaching performance parity on that hardware [1] and can now be run on this so it's actually sort of the least locked in framework. [1] It has been climbing the charts at https://github.com/soumith/convnet-benchmarks https://github.com/soumith/convnet-benchmarks for example.