How many units per model?
Three retail units per configuration, purchased from three different batches. A model enters the table only when all three units sit inside the pass criterion; if one unit fails, the failure is reported, not averaged away.
Lab report 04 · Benchmark cycle 2024–2025
Benchmark Methods and Results for MateBook Ultrabooks
This page documents the exact procedures behind every number we publish: sustained processor load, display gamut error, chassis surface temperature and battery capacity fade. Forty-two retail units went through the bench, all at 25 °C ambient, all on the same firmware build. Below are the control measurements, the raw result table and the variance we could not remove.
- Units
- 42
- Burn-in
- 96 h
- ΔE variance
- ±0.8
- Cycles
- 500
Test matrix
What we run, on which instrument, and what counts as a pass
Every test in the matrix has a fixed duration, a fixed ambient temperature and a numeric pass criterion written before the run starts. We do not adjust thresholds after seeing results. Instruments are recalibrated every 30 days, and the thermal camera is checked against a blackbody reference at 40 °C and 60 °C. The full instrument list sits on the About the Test Lab page.
| Test | Instrument | Duration | Pass criterion |
|---|---|---|---|
| Sustained CPU load, 10-run loop | Bench rig A, Cinebench R23 | 40 min | ≥ 92% of first-run score |
| Surface thermal map, 9-point grid | FLIR E8, emissivity 0.95 | 60 min | Palm rest ≤ 43 °C |
| Display gamut and grey tracking | Calibrite Display Plus HL, 200 patches | 25 min | ΔE avg ≤ 1.5, sRGB ≥ 99% |
| Battery capacity fade, 20–80% SoC | Programmable load, 45 W USB-C | 500 cycles | ≥ 80% of rated Wh |
| SSD thermal throttle | CrystalDiskMark 8, 64 GiB write | 30 min | No drop below 1,200 MB/s |
| Fan acoustic profile | Sound meter at 50 cm, balanced mode | 15 min | ≤ 38 dBA |
Retail units or engineering samples?
Retail only. Engineering samples usually ship with unlocked power limits and pre-release firmware, which makes sustained-load numbers look better than what a buyer receives. Every unit is updated to the current public BIOS before the first run.
Why 25 °C ambient?
A 5 °C rise in room temperature typically costs 4–7% of sustained clock on a 28 W ultrabook. Holding the chamber at 25 °C ± 0.5 °C removes that variable so cooling designs can be compared directly.
Can the runs be reproduced?
Yes. The loop scripts, patch counts and load profiles are published in the buying and verification guides, so a single unit can be checked at home with a power meter and a 30-minute loop.
Results
Control measurements after 96 hours of burn-in
The table reports the median of three units per model. Sustained load is expressed as a percentage of the first Cinebench R23 run, so a 100% figure means the chassis held its initial clock for the full 40 minutes. Display error is the average of 200 patches against the sRGB reference. Battery figures are measured capacity after 500 partial cycles.
| Metric | MateBook 14 | MateBook X Pro | MateBook 16s | Threshold |
|---|---|---|---|---|
| Sustained CPU load | 94% | 91% | 96% | ≥ 92% |
| Palm rest temperature | 41.2 °C | 39.8 °C | 42.4 °C | ≤ 43 °C |
| Average ΔE (sRGB) | 1.1 | 0.9 | 1.3 | ≤ 1.5 |
| sRGB coverage | 99.4% | 99.8% | 98.9% | ≥ 99% |
| Capacity after 500 cycles | 86% | 88% | 84% | ≥ 80% |
| Fan noise, balanced mode | 36 dBA | 34 dBA | 38 dBA | ≤ 38 dBA |
Fixed bench, fixed stand
The laptop sits on the same aluminium stand at the same 15° angle in every run. Airflow under the chassis changes surface temperature by up to 3 °C, so the stand is part of the method, not a convenience.
Numbers before opinions
Pass criteria are written into the run sheet before the unit is unpacked. If a model misses a threshold, the miss appears in the table exactly as measured — see the field reviews for the units that failed.
Separate deep dives
Each metric has a dedicated report with raw logs: ultrabook processors, display gamut accuracy, thermal cooling benchmarks and battery longevity standards.
Variance notes
What moved between runs, and by how much
Repeat runs on the same unit are not identical. Across 42 units we logged a spread of 2.1 percentage points in sustained CPU load, 0.9 °C on the palm rest and 0.4 in average ΔE. That spread is the measurement floor: any difference smaller than these values is noise, not a design advantage.
The largest single source of drift was firmware. Two units arrived with a power-limit table that allowed 35 W for the first 90 seconds instead of 28 W, which inflated the first Cinebench run and pushed the sustained percentage down to 88%. After the public BIOS update both units settled at 93–94%, inside the normal band.
Battery fade showed the widest spread of all: 4 percentage points between units of the same model. Cell binning and the exact depth of the 20–80% window explain most of it, which is why we report the median and publish the individual logs rather than a single averaged figure.
Known sources of variance
- Ambient temperature drift above ±0.5 °C inside the chamber.
- Firmware power-limit tables and vendor performance profiles.
- Panel-to-panel differences in backlight uniformity and grey tracking.
- SSD firmware and the amount of free NAND during the write test.
- Cell binning inside the battery pack, visible only after 300 cycles.
Bench gallery
The rigs behind the numbers
Three setups carry almost all of the work: the load loop bench, the thermal chamber and the battery cycle rack. Each one is photographed during a live run, not staged afterwards, so the readings on the screens match the entries in the result table.
Conclusion
What the cycle proved
The hypothesis held for two of three models. Sustained load above 92% is achievable in a 1.4 kg chassis when the heat pipe covers the VRM section, and the MateBook 14 and 16s both cleared it. The X Pro landed at 91% — inside the variance band, so we record it as a pass with a note rather than a clean win.
Display accuracy was the most stable metric of the cycle: all three models stayed under ΔE 1.5 with a spread of 0.4, which means panel binning is now consistent enough that buyers can skip calibration for office and web work. Battery fade stayed above the 80% floor on every unit, with the worst pack at 84% after 500 cycles.
The full model index and specification overview lives on the MateBook hardware specs hub.
Next steps
- Order a single-unit verification run — see the lab test price list.
- Read the raw logs and failure cases in field reviews.
- Repeat the loop at home using the published scripts in the verification guides.