Nvidia bans using translation layers for CUDA software to run on other chips

Article note: From the people who brought you the legal/artificial "You can't use consumer GPUs in data-centers" license restriction, a legal/artificial "You can't port our stack to other hardware, or even inspect it's behavior." license restriction. Nvidia must be getting nervous about their moat shrinking as people move to higher level languages for most accelerator-driven code. As much as the solution is "Keep getting off of CUDA," it's still a dick move.