Yeah no one really cares about this for large models. Iteration speed matters, and no one is waiting 50 times as long to train a LLM. Intel has always publishe

Yeah no one really cares about this for large models. Iteration speed matters, and no one is waiting 50 times as long to train a LLM. Intel has always published the "training per dollar" because no one else competes. Even for fine tuning you are almost always better off getting smaller GPU cloud instances.