In-depth observation of China's materialistic chip industry
Just over half a year later, we have come to a world of M4 chips.
Something went wrong.
The first is that time is too fast. In October last year, Apple's M3 chip was just launched. In May of this year, the M4 was launched. Pinching the fingers, there is only half a year between the two generations of main chips.
The second is who uses it. The M4 chip is installed on the new iPad Pro, not on the laptop.
iPad Pro/Source: Visual China
Here's the problem. Why "overthrow" M3 so quickly? Why isn't the new Mac coming with an M4?
What is Apple's calculation?
Quickly overthrow the M3
Every year, new iPhones and new Macs will be equipped with new chips - once a year, which is the inherent rhythm of Apple's main chips. But this rhythm was completely interrupted by the M4.
Starting in 2023, Apple will mainly promote 3nm chips, including the M3 and A17 Pro, which are manufactured by TSMC.
The M4 announced in 2024 has several promotional "highlights" in performance parameters.
For example, it has 28 billion transistors, which is more than the number of M3s; The second generation of 3nm technology "N3E" process is used instead of "N3B" adopted by M3; The CPU has 10 cores, including 4 performance cores and 6 efficiency cores, which is two more efficiency cores than the M3.
For example, M4 uses LPDDR5X to achieve a unified memory bandwidth of 120GB/s.
For example, M4 can handle 38 TOPS per second. But it's still lower than the Snapdragon X Elite's 45TOPS.
Apple, for example, claims that the M4 can compete with all the chips of today's laptops, without lagging, but with only a quarter of the power.
The configuration of the Apple M4 chip / Source: Apple
It is worth noting that the comparison cannot be "misplaced". The M4 is currently a "basic model", and compared to the basic model of the M3, the data on paper has indeed improved. However, compared with the M4 and the M3 Ultra, M3 Pro, M3 Max and other high-end M3s, the former is obviously not as good as the latter.
And, just how big is the "leap" of the M4? Apple has been comparing the M4 to the M2 – the previous generation of the iPad Pro used the M2. Therefore, Apple started with two generations of products, skipping M3 and relying on M4 to beat M2 to produce a "leap". This is a victory without force.
Because the product lines are inherently different. The iPad Pro can't be compared with the MacBook Air in the first place, so it's also strange to compare the chip.
Unless, Apple doesn't want to make you compare the M3 and M4 carefully.
The M4 configuration of the iPad can directly isolate the subject from the background / Source: Apple
In fact, there are as many GPU cores as M4 as there are M3 GPU cores. The CPU has two more cores, which is said to reduce the power, but the increase in these two cores may just be a buff for "stacking" to prevent M4 from running slower than M3.
The key lies in the difference between TSMC N3B and N3E. Although the N3E is the latest technology, it has a lower process density than the previous generation N3B - new but not strong, and Apple has little confidence.
TSMC's nightmare
On the N3 node, TSMC is a nightmare.
N3 is a 3nm FinFET (N3) process technology. The original version was the N3B, which had about 25 EUV layers, which is twice as many as the 5nm process. The N3E and N5 process nodes share the same SRAM (Static Random Access Memory) size, but with fewer EUV exposures.
The increase in N3B process density is slightly higher than the increase in wafer cost. After one operation, the cost per transistor was reduced by 11% – the weakest "progress" in major process technology in more than 50 years. And, on a per-transistor basis, the cost of everything except for speed improvements is either flat or negative, causing every buyer to be discouraged.
台积电在2022年曾讨论了四种 N3 衍生制造工艺——N3E、N3P、N3S 和 N3X/图源:gsmarena
N3B is expensive, and the "improvements" in performance, power, and density are lackluster, and can only end in "failure".
The failure of N3B is like the failure of TSMC's 20nm process. At that time, Qualcomm Snapdragon 810, Apple A8, Nvidia Tegra X1, MediaTek Helio X20, etc. all used TSMC's 20nm process, and all of them suffered from serious heating and runaway power consumption - TSMC has since been named "Taiwan leakage".
Anyway, in 2015, the big customers either quickly moved to the 16nm node, or they were reluctant to use the 28nm process. It was TSMC's darkest moment.
According to WikiChip, the N3 node seems to be "a one-time node", the kind that will not do it after completing this order.
So why does TSMC still do N3B? The main reason is that it has to deliver on its promise to early adopters of technology like Apple.
Except for Apple's M3 and A17 Pro, no one uses the N3B. Products such as Zen 5, Intel GPUs, and Broadcom custom ASICs either continue to use N5 or simply turn to N3E, a follow-up improvement process.
iPhone in Apple's flagship store/Source: Visual China
The pace of N3B development was chaotic, and TSMC missed the two-year upgrade cycle of major process nodes. Production will start in the fourth quarter of N3B2022 and in mid-to-late N3E2023.
TSMC's rhythm is messed up, and Apple's cycle of releasing the "trump card" is naturally messed up.
Moreover, a 7.3-magnitude earthquake occurred in Hualien County, Taiwan, China in early April, which affected the normal production of chips. The N3 wafer fab in Tainan suffered structural damage and the beams and columns were fractured. Cracks in the walls of the R&D lab are clearly visible.
For Apple, the core reason for pushing the M4 on the iPad first and not pushing the MacBook Air is that the N3E has just left the factory and the production is low. The sales of the iPad are much lower than that of the MacBook Air, and relying on the former to blow a wave of M4 will not create the problem of "delivery can't keep up".
A new showdown for AI PC
Apple is in a hurry to throw out the M4 and hit the "neural engine", obviously more concerned about the new outlet of artificial intelligence.
Giants such as Apple are also afraid of being abandoned by the "times".
Since 2023, Apple's computer and mobile phone sales have continued to decline. Since the beginning of this year, the stock price has begun to decline, and Buffett has also shorted a wave.
Since the beginning of 2023, Macs have been stumbling: sales have fallen by more than 20% year-on-year in almost every quarter, and even more than 30% year-on-year in the first quarter.
Apple's Q4 2023 earnings report, in which iPhone, Mac, and iPad sales are all declining
The iPhone has been the top smartphone seller in the world for many years. In the first quarter of 2024, this top spot was also taken away by Samsung. In the highly competitive Chinese market, Apple fell to third place, roughly on par with Huawei, which saw its sales grow nearly 70 percent, capturing some of the high-end markets dominated by Apple.
In January and February 2024, iPhone sales in China plummeted 37% year-on-year. In March, a large price reduction promotion was carried out, and finally barely pulled the sales decline in the first quarter back to 19%, which was the worst quarterly performance since 2020. The general environment is not good - in the first quarter of 2024, overseas smartphone shipments in China fell by 27% overall.
The fist products of the past are not selling well.
On the one hand, smart terminals really can't play any new tricks, and Apple's "squeezing toothpaste" type of innovation is nothing more than small repairs and small repairs, small fights, and few people pay the bill. On the one hand, AI laptops blessed by artificial intelligence have become a new outlet, and those that can't keep up will be eliminated.
Therefore, Apple urgently launched the M4 to send a clear signal to the market - "I am still the front of the AI PC line".
According to Tim Millet, Apple's vice president of platform architecture, the M4 delivers the same performance as the M2 with half the power. Compared to the new generation of PC chips of thin and light laptops, the M4 delivers the same performance with only 1/4 power consumption.
Advertisement for Apple M4 chip configuration
To make AI PCs, Apple's biggest enemy is Microsoft, so it hurriedly threw the M4 before Microsoft's development conference.
On May 20, Microsoft will host a Windows and Surface AI event in Seattle, where new Arm hardware is expected to be unveiled. Microsoft has said that its computer can beat Apple's MacBook Air (M3) in terms of CPU performance and AI-accelerated tasks.
At present, the definition of AI PC in the market is not clear. A more common concept is that AI PCs allow users to perform AI computing on their own devices without relying on the cloud, which is quite demanding on the chip.
Microsoft calls 2024 the "first year of AI PCs" and predicts that by 2026, half of commercial PCs will be AI PCs. Intel is also highly bullish on the market, and its CPU shipments of AI PCs in 2024 are expected to exceed the original target of 40 million.
Since Microsoft, which relies on OpenAI, has shouted "the first year of AI PC", Apple has to fight mentally.
After all, after ten years of building a car, once defeated. The original industrial design team also said goodbye. Losing the opportunity for generative AI, Apple has to face the "unbearable weight of life".
Author | Rong wisdom
Edit | to the by
Editor-in-Chief on Duty | Huang Mingting
Typography | Fifi