Apple's A18 chip NPU capability exceeds M4! However, the AI mobile phone market is surrounded by strong enemies

According to the latest reports from Taiwan media, as Apple's iPhone 16 series of mobile phones will be released this fall, as the date approaches, Apple began to increase the order size of the A18 processor. The supply chain revealed that the full range of iPhone 16 products is expected to be equipped with TSMC's second-generation 3nm process N3E. Due to Apple's market position, and considering the future cooperation between Apple and TSMC on nanoscale processes, the price of Apple processors on TSMC's N3E process has not changed in 2025. As a comparison, Nvidia has accepted TSMC's request for a price increase of about 10% for 4nm wafers in 2025.

According to the latest research from JPMorgan Chase, iPhone sales in China continued to grow in May, up 44% from April this year, thanks to the recovery of the Chinese market, of course. However, the global AI mobile phone market will flood into a number of powerful flagship products in the second half of this year, and the future of the iPhone 16 series is not entirely clear.

Apple's A18 provides strong power for the "AI version" of iOS

At this year's WWDC, Apple's senior vice president Craig Federighi announced in his speech that Apple Intelligence will reach a partnership with OpenAI's ChatGPT, and "hinted" shortly after the speech that Apple's iOS18 plans to introduce Google's Gemini platform this fall to provide users with a richer and smarter service experience. At present, this news has been mentioned by Mark Gurman, the whistleblower of Apple products, on the show.

Apple is also in discussions with Meta, Facebook's parent company, to consider integrating Meta's generative AI models into Apple Intelligence. Artificial intelligence startups Anthropic and Perplexity are also reportedly in talks with Apple. As Apple opens its own software ecological "wall", iOS's backward situation in AI has been completely shattered, and it has vaguely risen to become the leader of AI mobile phones, but this requires more powerful support from the A18 processor.

A number of Apple analysts said that the A18 processor may be the biggest update highlight of the iPhone 16 series of mobile phones, and the neural engine on the chip will surpass the Apple M4 chip in terms of computing power. On May 7, 2024, Apple officially released the M4 chip. Like the A18 processor, the M4 chip also uses TSMC's N3E process, with up to 10 cores and a unified memory bandwidth of 120GB/s. Based on the next-generation GPU architecture of the M3 series chips, Apple has created a new 10-core GPU for the M4 chip, bringing dynamic caching, hardware-accelerated ray tracing, and hardware-accelerated mesh shading to the iPad for the first time.

On the M4 chip, Apple paired it with the most powerful neural engine NPU to date, which is specifically designed to accelerate AI tasks. The NPU module can perform up to 38 trillion operations per second, which is up to 60 times faster than the original Neural Engine in the A11 Bionic chip. The M4 chip with a powerful NPU gives Apple the confidence to build an industry-leading AI PC product.

Now, Apple intends to further improve the NPU capability on the A18 processor, and its computing speed will most likely exceed 40 trillion times per second. However, even so, the iPhone 16 series does not dare to say that it is unrivaled in the AI mobile phone market, because the competitive strength of the opponents is also good.

The Android camp is highly competitive with AI processors

High-performance processors are at the heart of AI phones, and Apple understands this, as does the Android camp. In order to improve the AI capabilities of Android flagships, several powerful AI processors will also be unveiled during the same period as the iPhone 16 series.

The first is, of course, Qualcomm's Snapdragon 8 Gen4. Previously, Qualcomm announced that it would hold the Snapdragon Summit 2024 in Maui, Hawaii from October 21 to October 23, when Qualcomm's mobile platform Snapdragon 8 Gen4 will officially debut, and the Xiaomi 15 series will win the world premiere. Summarizing the leaked information, the Snapdragon 8 Gen4 processor may be codenamed "SUN" (sun), symbolizing its powerful performance and endless energy. In terms of chip design, the Snapdragon 8 Gen4 is based on TSMC's N3E process, equipped with a self-developed CPU large-core Oryon core, the CPU architecture is Nuvia Phoenix architecture, and the CPU frequency will reach a terrifying 4.2GHz, and in the Geekbench test, the single-core benchmark score reached 3000 points. The Snapdragon 8 Gen4's GPU architecture is said to be the "Slice GPU Architecture", which integrates the Adreno 830 GPU, which divides the GPU into multiple independent physical units, each with its own pipes, cache, and memory, allowing for more efficient task processing and resource allocation. According to tipsters, the biggest upgrade to the Snapdragon 8 Gen4 may come from the NPU. In the Snapdragon 8 Gen3, the performance of the Hexagon NPU is 98% higher than that of the Snapdragon 8 Gen2, and "it is expected that the NPU of the Snapdragon 8 Gen4 will continue to be multiplied."

Previously, the blogger "Digital Chatting Station" broke the news that the Qualcomm Snapdragon 8 Gen4 exceeded 10,000 points on the Geekbench multi-core score, which also surpassed Apple's A18 processor and created a new industry benchmark.

Of course, the Qualcomm Snapdragon 8 Gen4 is strong in performance and not low in price. According to South Korean media reports, due to the use of TSMC N3E process, coupled with a large number of self-developed architecture and self-developed cores, Qualcomm Snapdragon 8 Gen4 quotation skyrocketed by 25%, so Samsung plans to seek cooperation with MediaTek to let MediaTek customize Dimensity chips for Galaxy S25 and S25+, just using Snapdragon 8 Gen4 on S25 Ultra.

Able to compete with the Snapdragon 8 Gen4, MediaTek's product is the Dimensity 9400. Not long ago, the MediaTek Dimensity 9400 processor appeared in the IMEI database, and this processor platform will use the Arm public version of the Cortex-X925 CPU, as well as the Immortalis G925 GPU. At present, MediaTek's flagship processor has adopted an all-core architecture, taking Dimensity 9300+ as an example, this processor has 1 Cortex-X4 super core with a frequency of up to 3.4GHz, 3 Cortex-X4 super cores with a frequency of 2.85GHz, and 4 Cortex-A720 large cores with a frequency of 2.0GHz. It is expected that MediaTek will still use this architecture on the Dimensity 9400, and the super-large core will become a Cortex-X925 CPU, and the X925's single-core performance will be 36% higher than that of the X4 (based on Geekbench testing). Thanks to up to 3MB of private L2 cache, Arm says the X925 delivers a 41% improvement in AI workload performance (measured in "time to tokens"). Of course, MediaTek not only enhances the performance of the CPU and GPU, but also looks forward to the self-developed NPU core, the MediaTek NPU 790 on the Dimensity 9300+ The built-in hardware is MediaTek's seventh-generation NPU core, which is deeply adapted to the Transformer model to provide faster and safer edge AI performance, so the NPU performance on the Dimensity 9400 will not be bad.

In addition to the mainstream Snapdragon and Dimensity, the strength of the manufacturer's self-developed processor is also good. It is reported that the Tensor G5 chip, which is expected to be used in Google's flagship smartphone next year, will be based on TSMC's 3nm process and has successfully entered the tape-out stage. Past generations of Tensor processors can basically be understood as "variants" of Samsung processors, except for Google's own TPU (Neural Network Engine), which is basically modified and customized on Samsung Exynos. The Tensor G5 chip is a fully self-developed processor that uses completely independent chip design and TSMC's latest 3nm process. With the blessing of the Tensor G5 chip, Google's latest generation Pixel is also very worth looking forward to, especially in terms of AI capabilities, Google TPU can ensure that the latest AI technology in the Android system is supported by the fastest hardware.

If the project goes well, this list also includes Samsung's Exynos 2500, which we also covered in our previous article. The Samsung Exynos 2500 will be based on Samsung's second-generation SF3 process, using a "1+2+3+4" ten-core four-cluster CPU architecture, and will use Arm Cortex-X925 ultra-large core and Arm Cortex-A725 large core. However, the biggest uncertainty of this processor at present comes from Samsung's second-generation SF3 process, and the current yield is completely unable to support mass production.

epilogue

Apple's A18 is expected to bring more powerful NPU capabilities than the M4 chip, which is very necessary to enhance the AI capabilities of the iPhone 16 series. According to a research report, in the second half of this year, AI mobile phones are expected to drive consumer electronics components with both product volume and price, and the application of AI large models in mobile phones is expected to break the lack of innovation in the terminal market, and the iPhone 16 series is also expected to increase its market share.

However, in the AI mobile phone market, the prospects of the iPhone 16 series are not "stable", facing fierce market competition, Qualcomm Snapdragon 8 Gen4, MediaTek Dimensity 9400, and Google and Samsung's self-developed chips are expected to be significantly enhanced in AI processing power, which will have a certain impact on the iPhone 16 series equipped with A18 processors. Of course, consumers are happy about this competitive situation.