I think it started with CUDA. OpenCL never caught on as well as it needed to and compatibility layers for AMD cards had orders of magnitude worse performance. So, every time anyone wanted to do something fancy on a GPU, they bought NVidia.
Replies
Plenty of "woulda coulda shoulda" to go around -- the science community should have insisted on OpenCL or that CUDA itself be made open source. OpenCL devs should have zeroed in on performance and been able to show it beating CUDA in a bench or two. AMD shouldn't have fumbled compute for a decade.
sep 3, 2025, 9:37 am • 1 0 • view