There is probably some relevant information to other cards here but my benchmarks are on dual MI50 32GB cards because that is what I have, and thought I would share with the community. Install instructions at the end. I'll put a dump of the full llama-benchy tables in a comment in case anyone wants them, they include 3,4, and 8 concurrency levels, too (edit: it won't post the comment, maybe because my internet sucks, I'll try to later). For those that don't know, llama.cpp is available in the ...