After a earlier model of this information put face-to-face chips from Qualcomm, MediaTek, Apple, Samsung, and Huawei/HiSilicon, a bunch of issues modified within the smartphone part area, notably the arrival of Google, the demise and rebirth of flagship chips from Huawei on account of US sanctions, and Samsung getting back from a hiatus from the flagship area. First, let’s check out an general specs comparability of at present’s contestants.

Snapdragon vs. Dimensity vs. Exynos vs. Tensor vs. Apple: Specs In contrast

SoC components

Cellular system-on-chip (SoC) design has been fairly steady up to now few years, with most chip producers following an identical multi-core design with many specialised blocks devoted to offloading processing duties from the central processing unit (CPU):

• Graphics Processing Unit (GPU): Accountable for accelerating visible calculations, particularly 3D rendering.
• Picture Sign Processor (ISP): Course of knowledge captured from the digital camera sensors.
• Digital Sign Processor (DSP): Course of knowledge obtained from different sensors, typically employed for media and/or communication processing.
• Neural Processing Unit (NPU) (aka AI Accelerator): Specialised in synthetic intelligence and machine studying duties.
• Modem: Accountable for including help for mobile networks (e.g. 4G/LTE, 5G Sub-6 GHz, 5G mmWave…).

CPU cores are generally cut up between effectivity cores and efficiency cores, with the primary consuming significantly much less vitality however not notably quick. Heavier duties are offloaded when essential to the efficiency cores, with Android chips adopting a 3rd tier of even quicker CPUs denominated as “Prime cores”.

For nearly 10 years, all smartphones from conventional manufacturers have used the identical CPU structure: ARM. Nonetheless, not all ARM CPUs are equal. Apple, for instance, designs its personal CPU cores which can be appropriate with the ARM instruction set, whereas the Android suppliers undertake reference CPU cores from ARM, generally known as the Cortex cores.

With the success of Apple’s in-house designs, each Qualcomm and Samsung are anticipated to as soon as once more provide customized ARM designs, one thing that mirrors the early days of Android SoCs with Qualcomm Krait and Samsung Mongoose cores. As of 2024, ARM’s reference cores for cell processors are cut up between Cortex-X, Cortex A7xx, and Cortex-A5xx cores for prime, efficiency, and effectivity cores, respectively.

Previously three years, Android flagship SoCs had one predominant differentiating issue when in comparison with mid-range chips: The prime cores. Qualcomm’s launch of the Snapdragon 7+ Gen 2, nevertheless, made that distinction out of date, however that chip nonetheless contains a barely much less succesful GPU, ISP, and modem in comparison with the corporate’s flagship choices.

Manufacturing course of

Apple had the primary client processors made on TSMC’s 7-nm—technically introduced after a Huawei Kirin chip, however reaching shops first—, 5-nm, and 3-nm, the latter used for the present A17 processor used on the iPhone 15 Professional fashions. The identical privilege additionally applies to mid-cycle refreshes of these generations, such because the extra environment friendly N3E course of used on the A18 chips for the iPhone 16 household, adopted quickly after by the Dimensity 9400 chip. 

This new MediaTek flagship confirmed how far the partnership with its countrymen at TSMC advanced, as MediaTek’s chips had been historically one course of node behind Apple’s. Previously, the corporate was identified for not having energy-efficient processors, however that was influenced through the use of outdated fabrication processes from each TSMC and UMC, additionally from Taiwan. The Dimensity chips, however, are fabbed utilizing trendy processes, making them aggressive in vitality consumption.

Qualcomm, for its half, normally maintains a foundry-agnostic technique, with manufacturing booked in keeping with availability. The Snapdragon 888/888+ and eight Gen 1, for instance, had been made by Samsung Foundry, however their successor, the Snapdragon 8+ Gen 1 makes use of TSMC’s N4 course of, leading to much less energy consumption and warmth dissipation.

Samsung’s in-house chips (from its LSI division) are clearly made by Samsung Foundry, which for the previous few years has not been aggressive with TSMC’s nodes from the identical “nm” technology. The South Koreans are optimistic, nevertheless, that its 3 nm-class course of will provide a 45% enchancment in energy utilization, which might be a welcome evolution in cell phones.

As for Google, for the reason that Tensor cell chips are primarily based on Samsung LSI’s merchandise, its processors are additionally fabbed by Samsung Foundry, with rumors suggesting a future change to TSMC, time will inform. Tensor chips normally distinguish themselves from Exynos SoCs by using totally different CPU and GPU clusters, and most significantly, Google’s in-house Tensor NPU.

Huawei, for its half, is soldiering on within the SoC area, getting a lot of headlines for its “7nm” chip, made by Chinese language foundry SMIC. Nonetheless, the identical US sanctions that stopped them from contracting TSMC to make chips additionally block Huawei’s subsidiary HiSilicon from licensing trendy ARM CPU and GPU cores. Our exams with a Kirin 9000WL-powered system revealed a efficiency degree on par with mid-range chips just like the Snapdragon 7 collection or the Exynos 1480.

GPU tendencies

The primary one was AMD becoming a member of the cell SoC area with its Radeon GPUs on the Samsung Exynos 2200, utilizing the identical RDNA2 structure—though with far fewer cores—on the most recent PlayStation and Xbox consoles. Samsung’s retreat from the flagship cell SoC area, nevertheless, meant that it did not have a follow-up till two years later, with the Exynos 2400 chip used on the Galaxy S24 fashions packing an RDNA3 GPU.

Qualcomm’s Adreno GPUs observe their very own path, with only a few publicly accessible particulars with regards to specs. Current generations help ray tracing, alongside different superior options, making them additionally appropriate to be used in different Snapdragon chips designed for Home windows laptops.

MediaTek principally makes use of ARM’s reference GPUs in its Dimensity vary, with the newer Dimensity 9300/9400 powered by 12 cores of the RT-capable ARM Immortalis GPU. Previously, the Helio SoC household featured significantly fewer ARM Mali GPU cores than its Exynos and Kirin rivals, resulting in decrease efficiency, one thing that the flagship Dimensity fashions are altering due to a extra premium positioning and superior TSMC nodes.   

Google does not fairly match with different SoC manufacturers within the Android area, happening a special tempo on GPUs similar to on the CPU cores. The Tensor chips function a comparatively low seven GPU cores from ARM Mali household, so it’s incompatible with the identical options accessible on its rivals.

Apple does not disclose a lot about its in-house GPU designs, which share a whole lot of historical past with old-school PowerVR video processors. iPhone GPUs are nonetheless very succesful by way of options, sharing their core design with the bigger Apple M chips used on Mac PCs.

AI in every single place

With synthetic intelligence (AI) and machine studying (ML) much more within the highlight than in earlier years due to Secure Diffusion and ChatGPT, the significance of AI-dedicated cores is barely anticipated to extend. In that regard, flagship cell chips have been getting more and more extra highly effective in ML duties due to devoted cores akin to Apple’s Neural Engine, Qualcomm’s Hexagon, MediaTek’s NPU, and Google’s TPU.

Sadly, manufacturer-advertised numbers should not comparable between manufacturers, with Apple saying 16 trillion operations per second (TOPS) on its A16 Bionic chip, whereas Qualcomm’s numbers used to incorporate processing executed between the DSP, CPU, and GPU till the corporate stopped disclosing TOPS numbers for its Snapdragon chips.

Regardless, count on corporations to maintain promoting AI options on chips, throwing daring claims on efficiency enhancements, whereas working methods and apps attempt to catch up in supporting them.

Diminishing returns?