---
title: "Discussion on \"What 19 GB of Memory Compression Taught Me About MLX on M1 Max\" | Hashnode"
description: "Discussion on \"What 19 GB of Memory Compression Taught Me About MLX on M1 Max\". Disclaimer: This post reflects one solo operator's configuration on one M1 Max with 64 GB of unified memory in April 2026, running MLX + Qwen 3.6 35B-A3B Q8. Specific numbers (compressed GB, tok/s, wired limit) will differ on other hardware, other mo..."
url: "https://hashnode.com/posts/what-19-gb-of-memory-compression-taught-me-about-mlx-on-m1-max-1/69e73f126aa738a726bbf61c"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [What 19 GB of Memory Compression Taught Me About MLX on M1 Max](https://sleepyquant.hashnode.dev/what-19-gb-of-memory-compression-taught-me-about-mlx-on-m1-max-1?utm_source=hashnode&utm_medium=feed)

Disclaimer: This post reflects one solo operator's configuration on one M1 Max with 64 GB of unified memory in April 2026, running MLX + Qwen 3.6 35B-A3B Q8. Specific numbers (compressed GB, tok/s, wired limit) will differ on other hardware, other mo...

Apr 21 · 1 min read · sleepyquant.hashnode.dev  
Updated Apr 23  
Read the article

## Written by

[SleepyQuant](https://hashnode.com/@SleepyQuant)

@SleepyQuant · Ho Chi Minh City, Vietnam

  
[View profile](https://hashnode.com/@SleepyQuant)

Solo finance + tech enthusiast running a multi-agent quant on one M1 Max — all local, no cloud. Real trading numbers (including losses), MLX engineering notes, occasional existential thought. No signals, no hype. One honest email a week at sleepyquant.rest.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.