In September 2026, the mobile chip market is set for an unprecedented fierce battle. The Apple A20 Pro, MediaTek Dimensity 9600 Pro, Huawei Kirin 9050 Pro, and the recently launched Xiaomi Xuanjie O3 in August—these four chips are colliding head-on within the same timeframe. Interestingly, they have adopted four distinctly different technological approaches.
Apple A20 Pro
On the Geekbench benchmark website, the A20 Pro in the iPhone 18 Pro Max scored 4,727 in single-core and 12,424 in multi-core, with the single-core score directly breaking the historical record for mobile CPUs. Based on the 2nm process of TSMC (Taiwan Semiconductor Manufacturing Company), this chip features a 6-core CPU and 7-core GPU design, achieving a GPU score of 64,069, which even surpasses the iPad Pro equipped with the M4 chip.
What is truly intriguing is not the numbers themselves, but Apple's "conservatism." Officially claiming a 20% performance increase, actual tests show an approximately 25% improvement over the A19 Pro, with the single-core frequency pushed to 4.93GHz, approaching the 5GHz threshold. This strategy of "under-promising and over-delivering" is not rare in Apple's history, but placed at the debut node of the 2nm process, it feels more like a sense of ease—after all, with a single-core performance leading the Android camp by about 38%, there is indeed no rival that can shake its position in the short term.
However, Apple also has its own concerns. Although the multi-core score is impressive, Xiaomi's Xuanjie O3, using the mature 3nm process and relying on a higher core count, has managed to overtake it in multi-core benchmarks. This indicates that in terms of absolute multi-threaded throughput, the number of cores remains a hard threshold.
Dimensity 9600 Pro
The Dimensity 9600 Pro also adopts TSMC's 2nm process, but MediaTek has chosen a different path—instead of competing in single-core burst power, it pushes multi-core efficiency and cache capacity to the extreme.
Featuring an all-big-core architecture, it includes 2 C2-Ultra super cores at 4.55GHz, 3 C2 Pro big cores at 4.35GHz, and 3 C2 Pro big cores at 3.10GHz, with a total cache reaching an astonishing 34.5MB. MediaTek's logic is clear: with a sufficiently large cache, the latency of data movement is lower, and the efficiency of multi-core collaboration is higher. Official data also confirms this—a 15% increase in multi-core performance, while multi-core power consumption has actually dropped by 61%.
On the GPU front, the Mali G2-Ultra NX MC12 is equipped with a third-generation ray tracing engine, taking the lead in implementing the Raytracing Pipeline and hardware OMM ray tracing technology. Peak performance increases by 27%, and power consumption drops by 24%. This combination of moves is all about "energy efficiency."
Xuanjie O3
The Xuanjie O3 adopts TSMC's 3nm process but forcefully bridges the process gap through the aesthetic of brutal architectural scale.
The CPU specifications of this chip can be described as crazy: a 10-core all-big-core architecture, comprising 2 C1-Ultra super cores at 4.35GHz, 4 C1-Premium super cores, and 4 C1-Pro big cores, eliminating all low-power small cores. The Geekbench 6 multi-core score reaches 15,221, which is even higher than Apple's newly released A20 Pro, making its multi-core performance lead the other three by a wide margin.
Regarding the GPU, the Xuanjie O3 debuts the 16-core G2-Ultra NX, with performance increasing by 85% and ray tracing performance improving by 182%, while power consumption is reduced by 64%. More importantly, it is the world's first mobile processor to support LPDDR6, with a memory bandwidth reaching 113.8GB/s and memory access latency as low as 82ns.
Xiaomi's strategy is very clear: if the process cannot catch up, stack on core scale and memory bandwidth. The price of an all-big-core architecture is more difficult power control under daily light loads, but Xiaomi claims that in the 80% medium-to-low load range, the power consumption of the O3 is reduced by 25% compared to the previous generation.
Kirin 9050 Pro
If only looking at Geekbench scores, the Kirin 9050 Pro seems "not competitive." In the Zhuoyitong translation environment, it scores 1,539 in single-core and 5,295 in multi-core, which is not on the same order of magnitude as the other three. However, Huawei never intended to go head-to-head with Apple, MediaTek, and Xiaomi in this dimension.
The core keyword for the Kirin 9050 Pro is "Logic Folding." Traditional chip design lays out transistors on a 2D plane; the more advanced the process, the more transistors can be laid out. Huawei's approach is to arrange logic units in layers within a single chip, like converting a single-story house into a duplex, adding vertical interconnect channels inside. The result is: transistor density leaps from 155 million per square millimeter to 238 million per square millimeter, achieving a density increase within one generation that previously required three years of process iteration.
This means that without the most advanced lithography machines, Huawei bypasses process limitations through architectural innovation. Actual test data shows that under aligned performance conditions, the NPU power consumption of the Kirin 9050 Pro is reduced by 66%, GPU power consumption drops by 58%, and CPU performance core power consumption decreases by 41%.
Conclusion
The chip world war features four paths and four logics. Apple takes the "process-first" route, reaping the 2nm dividend and pushing single-core performance to the physical limit to serve the smooth experience of the iOS ecosystem. MediaTek takes the "energy-efficiency-first" route, trading large caches and an all-big-core architecture for multi-core efficiency and power consumption advantages. Xiaomi takes the "scale-first" route, using a ten-core all-big-core and LPDDR6 debut combination under the 3nm process to forcefully surpass Apple and MediaTek in benchmarks. Huawei takes the "architecture-first" route, using logic folding technology to trade for leaps in density and energy efficiency under process constraints. There is no absolute superiority or inferiority among the four routes; who will laugh in the end depends on the votes cast by users through their experience in real-world scenarios.