Ultrabook Processors: What Sustained Benchmarks Actually Show

Section 1 / Observations

Processor Overview: How Ultrabook CPUs Behave Under Sustained Load

An ultrabook processor is defined less by its peak boost clock and more by the power budget the chassis can actually dissipate. The same Core i7-1360P will hold 28 W indefinitely inside a 14-inch body with two heat pipes and a 70 mm fan, then collapse to roughly 15 W in a 13-inch shell with a single pipe. Our lab therefore records two numbers for every unit: PL1, the sustained package power, and PL2, the short-term boost power, together with the tau window the firmware allows before dropping back to PL1.

We tested nine configurations across three chassis sizes. Each unit ran Cinebench R23 multi-core for thirty minutes while package power, core temperature and effective clock were sampled every 500 ms. The gap between minute one and minute thirty is the most useful single indicator of real behaviour: a laptop that scores 12,000 points at the start and 8,400 at the end will feel noticeably slower in a long compile than its launch review suggests. The complete dataset is published on the benchmark methods and results page.

Core configuration and boost behaviour

Hybrid designs split work between performance and efficiency cores, and the scheduler decides which core receives a thread based on load history. In short bursts the arrangement is invisible to the user. Under a fifteen-minute export it matters: once the P-cores are power-capped, the scheduler moves background threads to E-cores, and measured package power settles about 3 W lower than the first-minute peak.

• 01PL1 (sustained): the power the cooling system can remove indefinitely. This is the number that governs long renders.
• 02PL2 (boost): the short-term ceiling, typically 1.8–2.5× PL1, limited by a tau window of 28–56 seconds.
• 03Tau: the timer that decides how long boost lasts. Vendor tools often let the user extend it at the cost of temperature headroom.
• 04Memory type: LPDDR5-5200 soldered on board versus DDR5 SO-DIMM changes both bandwidth and idle draw.

What the specification sheet leaves out

Retail listings quote the maximum turbo frequency and the base TDP, but never the firmware profile the manufacturer shipped. Two laptops with identical CPU model numbers can differ by 40 percent in sustained multi-core score. We also found that single-channel memory configurations cost 11–14 percent in integrated-graphics workloads and roughly 6 percent in compression tasks, which is why every unit in our sample was verified for channel count before testing. Related measurements for the panel and the battery are collected on the display gamut accuracy and battery longevity standards pages.

Section 4 / Thermal notes

Throttling Thresholds and What They Mean in Daily Use

Every unit in the sample began to reduce its package power once the internal diode reached 92–96 °C. The reduction is not a fault; it is the firmware protecting the die. What differs between devices is how quickly the limit is reached and how far the clock falls afterwards. The 13-inch single-fan unit hit its first reduction after 4 minutes 20 seconds and settled 31 percent below its peak clock. The 14-inch dual-fan unit reached the same limit after 11 minutes and settled only 14 percent below peak.

Fan behaviour matters as much as heatsink mass. Two of the tested devices kept the fan below 2,800 rpm for the first three minutes to stay quiet, which pushed the die past 90 °C almost immediately and forced an earlier power cut. A more aggressive curve starting at 3,400 rpm produced a 6 percent higher sustained score at the cost of roughly 4 dB(A) of noise. The full acoustic and surface-temperature data sits in the thermal cooling benchmarks report.

"The processor was never the bottleneck in our tests. The bottleneck was the rate at which the chassis could move heat away from the die — and that is a mechanical property, not a silicon one."

Practical consequences

• •Short tasks under 30 seconds run at boost power and are effectively limited by PL2, not by cooling.
• •Renders and compiles longer than five minutes settle at PL1; compare laptops by that figure, not by turbo clock.
• •Blocking the bottom intake with a blanket raises the die temperature by 7–9 °C within four minutes in our measurements.
• •Dust buildup of 0.3 mm on the fin stack reduced sustained score by 12 percent in a simulated twelve-month test.