# System Requirements for Running MTPLX on Apple Silicon: Hardware and Software Prerequisites

> Discover MTPLX system requirements for Apple Silicon. Get your Mac ready with macOS Sonoma, 16GB+ RAM, and Python 3.11+ for optimal performance. Learn hardware and software prerequisites.

- Repository: [Youssof Altoukhi/MTPLX](https://github.com/youssofal/MTPLX)
- Tags: getting-started
- Published: 2026-09-06

---

**MTPLX requires Apple Silicon (M1 or newer), macOS 14 Sonoma or later, 16 GB RAM minimum (32 GB recommended for 27B models), and Python 3.11+ for CLI usage.**

MTPLX is a native macOS application and command-line interface built specifically for Apple Silicon chips, leveraging Apple's **Metal framework** and **MLX library** for accelerated inference. Whether you're installing via Homebrew, pip, or the signed DMG, understanding the exact system requirements ensures optimal performance and prevents installation failures.

## Hardware Requirements

### Processor: Apple Silicon Only

MTPLX is architected exclusively for **Apple Silicon (M1, M2, M3, or newer)**. The runtime depends on Metal-accelerated kernels in `vllm_metal/metal/*` that target the unified memory architecture of Apple's ARM chips. Intel-based Macs are not supported.

The [`scripts/install_macos.sh`](https://github.com/youssofal/MTPLX/blob/main/scripts/install_macos.sh) installer performs an automatic hardware detection during onboarding. If non-Apple Silicon hardware is detected, the installation aborts immediately with a clear error message.

### Memory: Model-Dependent Minimums

| Model Size | Minimum RAM | Recommended RAM |
|------------|-------------|---------------|
| 4B and 9B | 16 GiB | 16 GiB |
| 27B "Optimized Speed" | 32 GiB | 48 GiB+ |

Memory pressure is the primary bottleneck for large model inference. MTPLX automatically detects available system memory and recommends a compatible model variant. Attempting to load a 27B FP16 checkpoint on a 16 GB machine will trigger an out-of-memory failure before inference begins.

### Disk Space

Allocate sufficient storage for your selected checkpoint:

- 4B/9B quantized models: ~15–25 GB
- 27B FP16 build: ~60 GB
- Virtual environment at `~/.mtplx`: ~2 GB additional

## Software Requirements

### Operating System: macOS 14 Sonoma or Later

MTPLX uses **Metal 3** features and kernel primitives introduced in macOS 14. Earlier macOS versions lack the necessary Metal driver support for MLX operations. Check your version before installation:

```bash
sw_vers -productVersion

```

### Python Runtime: 3.11 or Newer

Two installation paths exist with different Python requirements:

- **DMG installer**: Bundles its own isolated Python environment; no system Python required
- **pip/Homebrew installation**: Requires system-wide Python 3.11+

Verify your Python version:

```bash
python3 --version

```

The [`pyproject.toml`](https://github.com/youssofal/MTPLX/blob/main/pyproject.toml) file in the repository root declares `requires-python = ">=3.11"` and pins `mlx>=0.13` as a core dependency.

### MLX Support

No manual MLX compilation is required. MTPLX installs the stock **PyPI `mlx` package**, which includes prebuilt Metal drivers. The library automatically detects your chip generation and selects appropriate compute kernels:

- **M1/M2**: FP16-optimized builds
- **M3/M4**: FP8-native paths where available

## Optional: Thermal Management Tools

For sustained high-throughput workloads, MTPLX supports optional fan control integration:

```bash

# Enable maximum cooling profile (requires ThermalForge or TG Pro)

mtplx max

```

These tools are **not required for basic operation** but prevent thermal throttling during extended benchmark runs or batch inference.

## Installation Verification

After satisfying all requirements, verify your setup:

```bash

# Install via Homebrew (recommended)

brew install youssofal/mtplx/mtplx

# Or install via pip

python3 -m pip install -U mtplx

# Run system diagnostics

mtplx doctor

# Launch interactive CLI with auto-detection

mtplx start

```

The `mtplx doctor` command validates Apple Silicon detection, Metal driver status, available RAM, and model compatibility in a single report.

## GUI Installation Alternative

For users preferring a graphical interface:

```bash

# Download signed DMG

curl -L https://mtplx.com/download -o MTPLX.dmg

# Mount and install

open MTPLX.dmg

# Drag to /Applications; hardware check runs automatically on first launch

```

## Summary

- **Apple Silicon (M1+)** is mandatory—Metal acceleration requires unified memory architecture
- **macOS 14 Sonoma minimum** for Metal 3 kernel support
- **16 GB RAM baseline**, 32 GB+ for 27B models
- **Python 3.11+** for CLI installs; DMG bundles its own runtime
- Automatic hardware detection in [`scripts/install_macos.sh`](https://github.com/youssofal/MTPLX/blob/main/scripts/install_macos.sh) prevents incompatible installations
- MLX ships via PyPI—no manual driver installation needed

## Frequently Asked Questions

### Does MTPLX run on Intel Macs?

No. MTPLX requires Apple Silicon and the Metal framework. The [`install_macos.sh`](https://github.com/youssofal/MTPLX/blob/main/install_macos.sh) script aborts immediately on Intel hardware detection.

### Can I run MTPLX on macOS Ventura (13.x)?

No. MTPLX requires macOS 14 Sonoma or later for Metal 3 kernel features. Earlier versions lack the necessary driver support.

### How much RAM do I actually need for the 27B model?

The README specifies 32 GiB minimum, but 48 GiB or more is recommended for the "Optimized Speed" FP16 variant. MTPLX will recommend a smaller quantized model if your system has insufficient memory.

### Do I need to install MLX separately?

No. Running `pip install mtplx` or `brew install mtplx` pulls the correct `mlx` package from PyPI automatically. The Metal drivers are included with macOS 14+.