OpenCL did (and still does I think) work on nvidia cards. People I talked to back in the day complained more about OpenCL being "C but on GPUs" while cuda was m

OpenCL did (and still does I think) work on nvidia cards. People I talked to back in the day complained more about OpenCL being "C but on GPUs" while cuda was more akin to C++. They could move faster and do more in cuda and the nvidia lock in didn't matter as the fastest cards of the day were nvidia. I think vega cards were faster (or faster per dollar maybe) for some of the code that was relevant, but not by much and by that point legacy code lock in had taken over.