Home/All charts/Phone CPU Multi-Core Efficiency
Geekbench multi-core · Teardown-tested · 35 Chips · 2 test suites

Geekerwan Phone CPU Multi-Core Efficiency Curves

Geekerwan phone CPU multi-core efficiency curves cover 35 chips, with Geekbench 6 and Geekbench 7 charted separately. Phone CPUs spend most of their multi-core time between 2W and 6W; the top three and the peak champion quoted below come from Geekbench 6: under that range, the three phone CPUs scoring highest at the same board power are XRing O3, Exynos 2600, Dimensity 9600 Pro; looking only at the ceiling, the strongest peak belongs to XRing O3.

Updated 2026-11-09View source data ↗
PEAK · Geekbench 615013.95XRing O3 @ 20.97 W

Geekerwan Phone CPU Multi-Core Efficiency Curves

Geekbench 634 curves · axis 0–24.91W
Performance (points)Board power (W)02000400060008000100001200014000051015204W
How to read this chart
One chip, several versionsIf a chip ships in several memory or frequency versions it gets several separate curves (XRing O3 comes in LPD6 and LPD5X). The two tables below merge them into one row and keep the higher score, so one chip never takes two places.
Vertical dashed lineThe equal-power comparison point of the current suite is 4W — the best-covered of the 2/4/6W options across both suites, and inside the 2–6W range phone multi-core workloads actually use. The equal-power ranking reads this line, while the peak ranking reads the right end of each curve. Switching suites swaps both tables to that suite.
Very short curvesSome models only publish 2–3 frequency points upstream. Only those measured points are drawn, with no line between them, because the gaps were never measured. Models whose range does not cover 4W show “—” in the equal-power ranking but still count in the peak ranking.
Per wattPeak score ÷ peak power — how much score one watt buys at the peak point. Click the headers of the peak ranking to sort by score or by per-watt score.
Hover to readOnly the measured dots drawn separately on the chart can be read: they show the power and score actually tested. Continuous curves are fitted upstream and have no per-point measurements, so hovering them shows no reading — a fitted value is never presented as measured.
The two suites are not comparableGeekbench 6 and 7 changed both the workload and the scoring: the same chip usually scores lower on GB7, and the numbers cannot be converted. Switching suites also rescales the axes, so read the ticks.
2–6W is the range that mattersPhone multi-core workloads mostly sit between 2W and 6W, and that section is closest to real use. The part above 10W usually appears only for a moment during boot, a cold app launch, or a burst that needs fast response. Read the peak, but do not treat 10W+ scores as everyday performance.
The limits of a fixed-clock testDrawing a multi-core curve with a fixed-clock benchmark is a necessary compromise for testing: every core is pinned at each frequency step against the same load, which is very different from real use, where only one or two cores are usually active. Use the curve to compare total throughput at equal power; for everyday response and heat, lean on the per-core single-core efficiency curves.

The two tables below are derived from the upstream dataset (equal-power scores are interpolated along the upstream curves) and provided for reference only. If you believe something is off, email us via the address on the About page.

Equal-power efficiency (4W)

Geekbench 630 of 33 ranked4W board power · higher is better
#Chip4W
01XRing O3Xiaomi7665.12
02Exynos 2600Samsung6966.04
03Dimensity 9600 ProMediaTek6725.38
04Dimensity 9500MediaTek6543.21
05Snapdragon 8 Elite Gen 5Qualcomm6484.69
06XRing O1Xiaomi6367.53
07A20 ProApple6302.83
08SnapDragon 8 Elite Extreme Gen 6Qualcomm6287.29
09Snapdragon 8 EliteQualcomm6076.81
10A19 ProApple5777.84
11Dimensity 9400+MediaTek5691.86
12A19Apple5561.09
13Snapdragon 8 Gen 5Qualcomm5537.07
14A18Apple5477.23
15Tensor G6Google5420.89
16A18 ProApple5340.89
17Snapdragon 8s Gen 4Qualcomm5078.42
18Tensor G5Google5060.24
19Dimensity 9300+MediaTek5055.96
20A17 ProApple5049.74
21Snapdragon 8 Gen 3Qualcomm4971.09
22Kirin 9030 ProHuawei4877.37
23Snapdragon 8 Gen 2Qualcomm4530.34
24Snapdragon 8+ Gen 1Qualcomm4144.89
25Dimensity 9200+MediaTek4075.77
26Snapdragon 8s Gen 3Qualcomm4020.50
27Tensor G4Google3688.60
28Tensor G3Google3445.78
29Snapdragon 8 Gen 1Qualcomm3405.08
30Snapdragon 888Qualcomm3310.65
—A16 BionicApple—
—A15 BionicApple—
—Kirin 9020Huawei—

Full-load peak ranking

Geekbench 633 chipsDefault: peak score · headers switch to per wattPeak ↓
#ChipPeak power
01XRing O3Xiaomi15013.9520.97 W715.97
02Dimensity 9600 ProMediaTek13432.4123.23 W578.24
03SnapDragon 8 Elite Extreme Gen 6Qualcomm13233.9623.95 W552.57
04A20 ProApple13025.0013.50 W964.81
05Snapdragon 8 Elite Gen 5Qualcomm11986.9818.60 W644.46
06Exynos 2600Samsung11751.0516.16 W727.17
07A19 ProApple11111.0013.24 W839.20
08Dimensity 9500MediaTek10666.9317.61 W605.73
09A19Apple10554.0012.44 W848.39
10Snapdragon 8 Gen 5Qualcomm10280.4817.34 W592.88
11Snapdragon 8 EliteQualcomm10148.1115.26 W665.01
12XRing O1Xiaomi9919.9613.46 W737.00
13A18 ProApple9695.0011.86 W817.45
14A18Apple9640.0011.03 W873.98
15Dimensity 9400+MediaTek9305.7916.18 W575.14
16A17 ProApple8277.0010.04 W824.40
17Dimensity 9300+MediaTek8143.3212.37 W658.31
18Tensor G6Google7978.7410.96 W727.99
19Snapdragon 8 Gen 3Qualcomm7522.329.19 W818.53
20A16 BionicApple7368.008.50 W866.82
21Snapdragon 8s Gen 4Qualcomm7228.489.66 W748.29
22Tensor G5Google6705.788.97 W747.58
23Kirin 9030 ProHuawei6606.5510.55 W626.21
24A15 BionicApple6556.007.62 W860.37
25Snapdragon 8 Gen 2Qualcomm5691.096.80 W836.93
26Snapdragon 8s Gen 3Qualcomm5623.857.71 W729.42
27Kirin 9020Huawei5148.009.09 W566.34
28Dimensity 9200+MediaTek4959.137.55 W656.84
29Snapdragon 8+ Gen 1Qualcomm4704.685.75 W818.21
30Tensor G4Google4695.207.66 W612.95
31Tensor G3Google4388.376.81 W644.40
32Snapdragon 8 Gen 1Qualcomm4312.437.07 W609.96
33Snapdragon 888Qualcomm4102.576.91 W593.71

How to read these curves

What the two axes mean

The x-axis is “display-off board power”: the teardown runs on a dummy battery with the screen off, so idle board draw is only 0.05–0.1W and the number is essentially compute power (compute units and DRAM dominate). The y-axis is the Geekbench multi-core score. Each curve is one chip fitted from its lowest to its highest frequency: more power means more performance, and more heat and drain; the further a curve sits to the upper left, the stronger the multi-core performance — and usually the better the efficiency — at equal power, and the end of the curve is the peak.

Geekbench 6 or 7?

The two selectors on the left are “which workload” and “which chips”. Geekbench 6 has the widest coverage and is the one to use for comparing chips; Geekbench 7 is a heavier load with fewer chips measured, useful for seeing how new chips do on the current benchmark. Never mix the two.

Why 2–6W is what matters

Phone multi-core workloads mostly sit between 2W and 6W, so that is the section to read; above 10W usually appears only for a moment during boot, a cold app launch, or a burst that needs fast response. The two tables below rank the two scenarios — the equal-power ranking reads 4W, the peak ranking reads the ceiling. A higher ceiling does not mean a thirstier chip: XRing O3 draws 20.97W for 15013.95, while Dimensity 9600 Pro spends 23.23W for only 13432.41.

The limits of a fixed-clock test

A caveat: drawing a multi-core curve with a fixed-clock benchmark is a necessary compromise, and it differs a lot from real use, where only one or two cores usually work. Use it to compare total multi-core throughput at equal power; for everyday response and heat, lean on the per-core single-core efficiency curves.

Glossary

Board power (W)
Whole-board power measured with a dummy battery and the screen off. The test environment is clean: idle board draw is only 0.05–0.1W, so this is essentially compute power (compute units and DRAM dominate). Not battery output, not charging power.
Multi-core / single-core
Single-core measures one core’s responsiveness (app launches, taps); multi-core measures the total throughput of all cores working together (export, rendering, background work). Real use keeps only one or two cores busy most of the time, so read the multi-core curve together with the single-core ones. The single-core curves live on their own page.
Geekbench 6 / 7
Two generations of a cross-platform multi-core benchmark with different workloads and scoring: the same chip usually scores 5–10% lower on GB7, and the two sets cannot be converted.
Peak score / peak power
The highest point of a curve and the power it took. A multi-core full load pushes board power past 20W, but the part above 10W usually appears only during boot, a cold app launch or a fast-response burst, so it is not daily use.
Per watt
Peak score ÷ peak power. A different lens from the equal-power ranking: one is the cost of the ceiling, the other the output at a fixed wattage.
Sweet spot
Where the curve starts to steepen: another step in clock buys less score and costs more power. Read it from the bend of the curve in the current suite.
Fixed-clock curve
A power-performance line built by pinning the chip at each frequency step against the same load. It makes comparison easy, but real use changes core count and frequency constantly, so treat it as a reference rather than a real-world workload.

Multi-core efficiency FAQ

How is multi-core efficiency different from single-core?

+

Single-core measures one core’s ceiling, which decides tap and app-launch response; multi-core measures the total of all cores working together, which decides export, rendering and background work. Real use keeps only one or two cores busy most of the time, so lean on the single-core curves for daily experience — the two pages are not interchangeable.

Do these multi-core curves represent real-world use?

+

Not fully. Drawing a multi-core curve with a fixed-clock benchmark is a necessary compromise: it pins every core at each frequency step against the same load, while real use changes core count and frequency constantly. Use it to compare total multi-core throughput at equal power; for everyday response and heat, lean on the per-core single-core efficiency curves.

How should I read 2–6W versus above 10W?

+

2–6W is the range phone multi-core workloads actually use, and the one worth reading closely. Above 10W usually appears only for a moment during boot, a cold app launch or a burst that needs fast response. Read the peak, but do not treat 10W+ numbers as everyday performance.

Can I compare Geekbench 6 with 7?

+

No. Both the workload and the scoring changed, and the same chip usually scores 5–10% lower on GB7. Compare within one suite only.

More Geekerwan charts

Same chips, a different lens.