Even if it worked though, they're usually heavily bandwidth bottlenecked and near useless for LLM inference. CPU wins every time.
Even if it worked though, they're usually heavily bandwidth bottlenecked and near useless for LLM inference. CPU wins every time.