exelban
CPU/GPU/ANE Power Readings Doubled at ≥2s Update Interval
Issue:
When sensor update interval is set to 2 seconds or higher, reported CPU, GPU, and ANE power values are ~2x higher than actual (validated against powermetrics, asitop and physical USBC power meter). Total system power reading remains accurate.
Steps to reproduce:
- Set update interval ≥2s.
- Observe CPU/GPU/ANE power readings during heavy load (e.g., ANE reads ~12W vs. actual max ~6W. CPU reads ~100W vs actual ~50W)
- Compare with
powermetrics— readings match only when update interval in Stats is set to 1s.
I suspected this might be caused by the concurrent powermetrics process I used for validation, however the issue also persists in isolation.
Impact:
False/misleading power analysis for components.
Expected:
Accurate component power readings regardless of update interval.
Happy to help with testing etc
Setup
- M4 Max MacBook Pro
- MacOS Tahoe 26.1 Beta (25B5072a)
Screen recording of bug attached, showing discrepancy between Stats and asitop when Stats set to 2 second update interval:
https://github.com/user-attachments/assets/7a70d425-48ab-467f-a54d-02f89c2cf01e
Oh, that explains a lot. I was about to create an issue that the 'System Total' Power usage (not 'Average System Total') is sometimes consistently about half of either CPU or GPU power usage alone.
The update interval I have configured is 5sec.
M4 Tahoe 26.1
