ParentFull threadxiphias·AMD should just do it for TensorFlow. They would get a lot of benefit if they could show higher performance per dollar at least on Linux, and it would take just a small team to implement it.View on HN