>>106158578
>>106158595
i'm using a 3090, 3060 and some shitty 2666 ram to get to these speeds on q2xl on vanilla llama.cpp