-
-
Notifications
You must be signed in to change notification settings - Fork 1.1k
Description
Hi, first of all, thanks for Stats.
It’s an essential tool for understanding real-world workload behavior on Apple Silicon.
I wanted to start a discussion around Neural Engine (ANE) visibility, not as a concrete feature request yet, but more as a feasibility question.
Motivation
For example, when running local inference (Ollama, LM Studio, etc.), I consistently see:
GPU usage → ~100%
CPU usage → very low
Memory pressure → consistent with the loaded model
This matches expectations and is extremely useful feedback.
Question
Conceptually, the ANE is the third major domain on Apple Silicon (CPU / GPU / ANE). If workloads ever meaningfully hit these cores, having some visibility, similar to CPU/GPU, would help users reason.
From what I understand so far (please correct me if I’m wrong):
- The Neural Engine is real, dedicated silicon (not just marketing)
- macOS may not expose a meaningful system-wide utilization signal for ANE, unlike CPU/GPU
So the core question is:
Does ANE have any visibility like /proc/stats on linux?
Also, does the GPU stats right now bundle the ANE stats internally?
Thanks again for the project.
As you know M5 specs: 10-core CPU, 10-core GPU, 16-core Neural Engine