Apparently Itanium works quite well for LLMs https://medium.com/@tglozar/running-llama-inference-on-intel...

Which makes sense I suppose.