Oh yeah, same here. I'm not holding my breath, but I'll tell you what I'd really like: I'd really like to see Vulkan become the lingua franca of GPGPU compute. It's probably more possible for something like that to happen now than ever, since I think the majority of GPGPU usage is ML via PyTorch. That said, it's pretty obviously very different from CUDA, it feels unclear if this is really a promising path to go down.