Over/under on years to CUDA alternative? Thinking Obj-C/Swift style shift in programming ergonomics and optimizations to compiled version.

Farcaster Development Boutique @ https://dTech.vision 💜 in love with tech 🤓decentralization, fitness and oldschool rap 😎

I think NVIDIA should be the one driving it the most. They can't risk of losing devs on their platform due to sub-optimal programming interface.

who'd be the driving force behind it? 

seems like Intel + AMD would need to come together? maybe ARM + Apple based companies?

OpenAI’s Triton and PyTorch’s torch.compile have achieved this imo. They provide very convenient high level interfaces that get compiled down to highly optimized CUDA kernels for most use cases.