https://www.macrumors.com/2025/03/12/mac-studio-with-512gb-ram-delivery-delay
https://www.reddit.com/r/LocalLLaMA/comments/1j8r2nr/m3_ultra_512gb_does_18ts_with_deepseek_r1_671b_q4/
https://x.com/alexocheema/status/1899735281781411907
Whoever says Apple is not focused on AI, you are wrong.
Mac Studio M3 Ultra 512GB is the proof of that. A big leap from maximum 196GB RAM of M2 Ultra to 512GB is not common for Apple. Particularly when M3 Max has a maximum of 128GB RAM and Ultra version should just be a 2 Max chips packaged together, that’s maximum of 256GB RAM. Apple went out of its way to build this 512GB for a reason – running very large AI models.
Users are all excited to be able to deploy these model locally with the new 512GB machines. 18 Tokens per sec with Deepseek R1 671B Q4. Or have 2 of these 512GB beasts linked up with Thunderbolt 5 and run the full 671B model.
10k USD each seems expensive. But being able to run these models locally is exciting for many user. Just can’t easily get a PC with similar performance at this price range.
Leave a Comment