Mac Studio M3 Ultra LLM real world performance report

In this video I compile data from Reddit and some other sources, to show how fast the Mac Studio M3 Ultra will perform in real life.

This includes some surprises: in some cases the Mac Studio M2 Ultra will perform faster than the Mac Studio M3 Ultra!

Also a brief comparison with the Mac Studio M4 Max is included - if you are considering buying a Mac Studio for running LLMs on it, this video is a must watch to reset your expectations and understand a bit more of the background.

I introduce an alternative (by NVIDIA - RTX 6000 Pro Blackwell Generation), and also discuss Zeitverstärker, an LLM appliance based on top of the Mac Studio M3 Ultra, which you can purchase from one of my companies (buyzero.de).

Here are some useful links for you:

https://buyzero.de/products/zeitverst... - you can purchase the Zeitverstärker Mac Studio M3 Ultra LLM appliance here, pre-installed and custom-tailored to your demands. We are based in Germany, but ship world-wide.

https://www.nvidia.com/en-us/data-cen... - the RTX Pro 6000 will offer a significantly higher throughput, thanks to it's 1.6 TB/s data rate

https://github.com/ggml-org/llama.cpp... - performance data in token/s for different Apple Silicon Models. Comparing Llama 7B across the different models

  / some_base_mac_studio_m4_max_llm_and_comfyu...  

  / mac_speed_comparison_m2_ultra_vs_m3_ultra_...   - a wonderful post by SomeOddCodeGuy on Reddit, directly comparing M3 Ultra and M2 Ultra performance, with max. RAM respectively.

Share in the comments below, if you want me to share the Excel sheet I have created with the data compilation, and also of course if you have any questions, or are interested in a Zeitverstärker LLM appliance based on the Apple Mac Studio M3 Ultra.

Subscribe to my channel for more tips on AI for managers, entrepreneurs and business people, upcoming AI tools which will save you time and make you money.