Mac Studio (M2, 32GB unified memory)
- Power measurement
- wall-meter
- OS family
- macOS
- Chip
- Apple M2
- Unified memory
- 32GB
Confirm exact M2 variant (Max/Ultra) and macOS version before the first run.
Local AI Benchmark Dataset
This is the public home for measured local-AI benchmark rows from Max's owned rigs. The dataset is intentionally empty until real runs land; no estimated tokens/sec or invented power numbers are published here.
Schema version
1
Registered rigs
2
Measured rows
0
Rig registry
Mac Mini (M4)
- Power measurement
- wall-meter
- OS family
- macOS
- Chip
- Apple M4
Confirm unified memory size and macOS version before the first run.
Measured rows
No measured rows published yet.
This is deliberate. Rows will appear only after measured runs record the rig, model, quantization, context length, token counts, runtime, power-measurement point, and run date.
Methodology contract
- Discrete-GPU and Apple-Silicon rows stay separate and are never averaged together.
- Every row records prompt tokens, generation tokens, context length, quantization, runtime, OS version, and run date.
- Power rows must name their measurement point: wall meter, board sensor, powermetrics, or another explicit source.
- Article claims that cite benchmark rows must carry a matching
data-benchmark-idso the build gate can verify the row exists.
Dataset license
Benchmark rows and the JSON dataset download are published under theApache License 2.0. Article text, images, and product photography are not part of that dataset license.