FLOPS/$ is better on AMD afaik. double speed fp16 is only on gp100 and I don't know if any framework supports fp16 training yet. But yes AMD need to add 2x fp16

FLOPS/$ is better on AMD afaik. double speed fp16 is only on gp100 and I don't know if any framework supports fp16 training yet. But yes AMD need to add 2x fp16 too. The main reason people don't use AMD for deep learning is lack of fast libraries, and AMD's inability to provide a CuDNN equivalent optimised for their hardware