# DeepSeek (671B) running on a cluster of 8 Mac Mini Pros with 64GB RAM each

**URL:** <https://forum.devtalk.com/t/deepseek-671b-running-on-a-cluster-of-8-mac-mini-pros-with-64gb-ram-each/185709>\
**Category:** AI In The News\
**Tags:** ai, macs, deepseek\
**Created:** [29 January 2025 13:16 UTC](https://forum.devtalk.com/t/deepseek-671b-running-on-a-cluster-of-8-mac-mini-pros-with-64gb-ram-each/185709 "2025-01-29T13:16:26Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![AstonJ](https://forum.devtalk.com/user_avatar/forum.devtalk.com/astonj/32/7_2.png) [@AstonJ](https://forum.devtalk.com/u/AstonJ)\
**Post date:** [29 January 2025 13:16 UTC](https://forum.devtalk.com/t/deepseek-671b-running-on-a-cluster-of-8-mac-mini-pros-with-64gb-ram-each/185709/1 "2025-01-29T13:16:26Z")

</div>

This is cool!

> # DEEPSEEK-V3 ON M4 MAC: BLAZING FAST INFERENCE ON APPLE SILICON
> 
> We just witnessed something incredible: the largest open-source language model flexing its muscles on Apple Silicon. We’re talking about **the massive DeepSeek-V3 on M4 Mac, specifically the 671 billion parameter model running on a cluster of 8 M4 Pro Mac Minis with 64GB of RAM each – that’s a whopping 512GB of combined memory!**
> 
> This isn’t just about bragging rights. It opens up new possibilities for researchers, developers, and anyone interested in pushing the boundaries of AI. Let’s dive into the details and see why DeepSeek-V3 on M4 Mac is such a big deal.
> 
> ## TABLE OF CONTENTS
> 
> - [The Results Are In: DeepSeek V3 671B Performance on the M4 Mac Mini Cluster](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-the-results-are-in-deepseek-v3-671b-performance-on-the-m4-mac-mini-cluster)
> - [Why So Fast? Understanding the DeepSeek-V3 on M4 Mac Performance Advantage](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-why-so-fast-understanding-the-deepseek-v3-on-m4-mac-performance-advantage)
> - [LLM Inference Explained: A Systems Perspective on Running Large Models](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-llm-inference-explained-a-systems-perspective-on-running-large-models)
> - [Mixture-of-Experts (MoE) Models: The Key to DeepSeek V3’s Efficiency](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-mixture-of-experts-moe-models-the-key-to-deepseek-v3-s-efficiency)
> - [DeepSeek-V3 on M4 Mac: Why This Setup Works So Well for MoE Models](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-deepseek-v3-on-m4-mac-why-this-setup-works-so-well-for-moe-models)
> 
> - [Exploring Key Considerations: Power, Cost, and Alternative Setups for Running DeepSeek-V3](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-exploring-key-considerations-power-cost-and-alternative-setups-for-running-deepseek-v3)
> - [Power Consumption:](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-power-consumption)
> - [Efficiency:](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-efficiency)
> - [Older But Beefier Hardware:](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-older-but-beefier-hardware)
> 
> - [Conclusion: The Future of LLM Inference on Apple Silicon with DeepSeek-V3 on M4 Mac](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/#h-conclusion-the-future-of-llm-inference-on-apple-silicon-with-deepseek-v3-on-m4-mac)

> **[DeepSeek-V3 on M4 Mac: Blazing Fast Inference on Apple Silicon](https://digialps.com/deepseek-v3-on-m4-mac-blazing-fast-inference-on-apple-silicon/)**
>
> Discover how a cluster of M4 Mac Minis handled the enormous DeepSeek-V3 model. Performance benchmarks and insights inside.

---

<div class="post-metadata">

**Author:** ![AstonJ](https://forum.devtalk.com/user_avatar/forum.devtalk.com/astonj/32/7_2.png) [@AstonJ](https://forum.devtalk.com/u/AstonJ)\
**Post date:** [29 January 2025 18:38 UTC](https://forum.devtalk.com/t/deepseek-671b-running-on-a-cluster-of-8-mac-mini-pros-with-64gb-ram-each/185709/2 "2025-01-29T18:38:13Z")

</div>

> We just got the biggest open-source model running on Apple Silicon.
> 
> Without further ado, here are the results running DeepSeek v3 (671B) on a 8 x M4 Pro 64GB Mac Mini Cluster (512GB total memory):
> 
> | Model | Time-To-First-Token (TTFT) in seconds | Tokens-Per-Second (TPS) |
> | --- | --- | --- |
> | DeepSeek V3 671B (4-bit) | 2.91 | 5.37 |
> | Llama 3.1 405B (4-bit) | 29.71 | 0.88 |
> | Llama 3.3 70B (4-bit) | 3.14 | 3.89 |
> 
> Wait, Deepseek has 671B parameters and runs faster than Llama 70B?
> 
> Yes!
> 
> Let me explain…

> **[12 Days of EXO](https://blog.exolabs.net/day-2/)**
>
> 12 Days of Truly Open Innovation