I believe what they mentioned is that even though it works with Nvidia cards, Nvidia needs to make Driver optimizations to get good performance out of it, Which I believe if they do, they might be able to implement it better than AMD since they have extra Tensor and RT cores that they can...