If you own a Mac Studio, MacBook Pro, or Mac Pro with Apple Silicon, you have a genuinely capable GPU sitting in your machine. The M4 Max packs up to 128 GB of unified memory and a 40-core GPU that can handle real ML inference and training workloads. But try to monitor it, and you hit a wall.

The Activity Monitor Problem

macOS Activity Monitor shows you CPU and memory usage, but its GPU information is almost useless for serious work. You get a vague "GPU Usage" percentage with no breakdown of:

How much of the unified memory is allocated to the GPU vs. the CPU
Per-process GPU memory attribution
GPU utilization over time
Temperature or thermal state
Power consumption
Memory allocation trends (is usage growing?)

For anyone running ML inference with mlx, CoreML, or even PyTorch with MPS backend, this lack of visibility is a real problem. You are running workloads on a powerful GPU and have no idea what it is actually doing.

Why Apple Silicon is Different

Apple Silicon uses a unified memory architecture (UMA), meaning the CPU and GPU share the same physical memory pool. This is great for performance — no data copying between CPU and GPU — but it makes monitoring harder. Traditional GPU tools like nvidia-smi or nvitop do not work because Apple uses its own Metal and IOKit APIs instead of NVML or ROCm.

To get real metrics from Apple Silicon, you need to talk to IOReport — Apple's low-level performance counter framework. This is what gpulse does. It reads directly from IOKit and Metal to extract real GPU utilization, memory allocation, temperature, and power draw data.

What gpulse Shows You on Apple Silicon

Once you install gpulse, you get immediate access to metrics that macOS does not surface:

GPU utilization — actual compute utilization percentage, updated in real time
Memory usage — how much unified memory the GPU is using, with allocation trends
Temperature — GPU die temperature with color-coded thresholds
Power draw — current GPU power consumption in watts
Per-process attribution — which processes are using the GPU and how much memory each one has allocated
Leak detection — whether GPU memory usage is trending upward over time

All of this works through the Grid view, Detail view, and the Predict view for leak analysis.

No Other Tool Does This

As of today, gpulse is the only terminal-based GPU monitoring tool that supports Apple Silicon with full metrics. nvidia-smi is NVIDIA-only. nvitop requires NVML. btop shows minimal GPU data on macOS. If you have an M1, M2, M3, or M4 chip, gpulse is built for you.

Try It

Install takes 10 seconds:

brew tap gpulseai/gpulse && brew install gpulse

Then launch with gpulse dashboard. No configuration needed — gpulse auto-detects your Apple Silicon GPU at startup. See the getting started guide for the full walkthrough.