apple-smi 0.1.3

nvidia-smi equivalent for macOS Apple Silicon – Monitor Metal GPU usage, power, and temperature

🍎 apple-smi

nvidia-smi equivalent for Apple Silicon – High-performance, sudoless GPU monitoring with zero dependencies.

$ apple-smi
Wed Mar 04 14:52:00 2026
+-----------------------------------------------------------------------------------------+
| APPLE-SMI 0.1.3              macOS Version: 26.3 (25D125)              Metal Version: 3 |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                               |                 Disp.A |                      |
|      Temp                 Pwr:Usage/Cap |           Memory-Usage |             GPU-Util |
|=========================================+========================+======================|
|   0  Apple M1 (8-Core GPU)              |                     On |                      |
|       30C                    2.4W / 20W |    13826MiB / 16384MiB |                  12% |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
| GPU           PID  Type   Process name                                       GPU Memory |
|                                                                                   Usage |
|=========================================================================================|
|   0         72743    C    python3.11                                             530MiB |
|   0          1007    G    NotificationCenter                                     170MiB |
|   0          3197    G    WeType                                                 216MiB |
+-----------------------------------------------------------------------------------------+

✨ Features

• 🚫 No sudo required – Uses private macOS APIs (IOReport, IOHIDSensors, SMC) for sudoless operation.
• ⚡ Ultra-Fast & Lightweight – Optimized startup (<50ms) using targeted sysctl and ioreg calls.
• 🔋 Power Monitoring – Displays SoC power consumption (Usage) against chip TDP (Cap).
• 🌡️ Smart Thermals – High-accuracy GPU/SOC temperature readings with sensor fallback logic.
• 📋 Robust Process Listing – Lists active GPU processes with full name resolution (no truncation for paths with spaces).
• 🔍 C/G Classification – Categorizes processes by Type (C for Compute, G for Graphics).
• 💾 Memory Usage – Detailed breakdown of unified memory utilization.
• 🔄 Watch Compatible – Default 100ms sampling interval for smooth real-time monitoring.
• 📦 Zero Dependencies – Pure Python, utilizing ctypes to interface with macOS frameworks.

📥 Installation

pip install apple-smi

🚀 Usage

Command Line Options

# Basic usage
apple-smi

# Set sampling interval (default: 100ms)
apple-smi --interval 250

# JSON output for integration with other tools
apple-smi --json

# Continuous monitoring (highly recommended)
watch -n 0.1 apple-smi

Process Filtering

By default, apple-smi only shows Compute (C) processes (e.g., Python, MLX, Torch, llama.cpp) to reduce system noise. You can show all GPU-connected processes (including window compositors and browsers) using an environment variable:

APPLE_SMI_SHOW_ALL_PROCESSES=1 apple-smi

Backends

apple-smi automatically selects the best available backend:

1. IOKit (Default): Uses undocumented IOKit APIs for sudoless monitoring.
2. Powermetrics (Fallback): Uses the system powermetrics tool (requires sudo or root privileges).

Variable	Values	Description
APPLE_SMI_BACKEND	iokit, powermetrics	Force a specific backend. Default: auto-detect.
APPLE_SMI_SHOW_ALL_PROCESSES	0, 1	Show all GPU processes (G+C). Default: 0.

🔧 Requirements

• macOS on Apple Silicon (M1, M2, M3, M4, M5)
• Python 3.10+

📝 License

MIT License – see LICENSE for details.