I have no ideas about the complexity of the underlying CUDA/ROCm GPGPU, I am just using my card with CUDA a bit stupidly.
But why AMD do not setup a team of 5 good coders to have a good support of the well deployed/used frameworks in the happy path case? Later they can cover more cases but at least it would drive the adoption of their cards.
But why AMD do not setup a team of 5 good coders to have a good support of the well deployed/used frameworks in the happy path case? Later they can cover more cases but at least it would drive the adoption of their cards.