The unified memory architecture of M3 Ultra integrates the most high-bandwidth, low-latency memory ever available in a personal computer. Starting at 96GB, it can be configured up to 512GB, or over half a terabyte.
512GB of RAM. Holy crap that's a lot.
Reading on Hacker News and doing some research it seems like you need this much RAM to run certain open source LLMs locally. For example running the larger (in parameter size) Deepseek or Google Gemini models (via Ollama) would mean having a big powerful machine.