Comment by lukev

There's a massive financial incentive (billions) to allow existing CUDA code to run on non-NVidia hardware. Not saying it's easy, but is implementation difficulty really the blocker?