Recently, four domestic CPU manufacturers, including Loongson, Feiteng, Zhaoxin, and Kunpeng, held discussions, providing clear introductions to the development status, roadmaps, and goals of their CPUs. Although the presentations were relatively polite, there were still underlying tensions evident in their words. This is largely due to the fierce competition among these CPUs in the information technology sector, and the differing paths each company is taking. Once the higher-ups determine a “great path” or the main direction of the “great path,” the other CPUs’ market support from relevant entities would be severely jeopardized, akin to a disaster. Below, let’s review the paths of these four CPU companies.
Loongson’s Path: Starting Anew with Independence
Loongson’s path can be summarized as starting anew, being independent, and self-reliant.
Technologically, there should be no reliance on foreigners, and there should not be a constant expectation to bring source code back from abroad. The development roadmap should be independent and self-reliant, constructing a system independent of Wintel and AA, achieving a third of the world for itself. Loongson also proposed that the essence of the IT industry is “solution-oriented,” where the solution determines the dominance.
Loongson’s PPT pointed out that in the computer field, Intel makes more money than the entire machine manufacturers, while in the mobile field, Apple earns more than chip manufacturers, the root cause lies in industry dominance.
The computer system is dominated by Wintel, and PC manufacturers have no say; in the mobile system, ARM and Google or Apple dominate, while domestic manufacturers like HOVM and SoC manufacturers like Zhanrui have no say. This lack of say directly reflects the lack of discourse power, only able to earn meager profits within the industry division defined by foreign companies.
Mastering core technology is just the first step; industry dominance is fundamental. If one cannot grasp dominance, it will lead to becoming a tenant of foreign companies, working tirelessly to create high profits for them while only earning meager wages.
This state cannot support the long-term development of the industry. Mastering core technology is just the first step; industry dominance is fundamental. The fundamental way out for China’s IT industry lies in establishing its own ecosystem and constructing an independent IT technology system and industrial ecology.
Loongson’s PPT is the most detailed among the four companies, with relatively mature theoretical thoughts, and some data is quite solid, even revealing how many chips were sold. Below are parts of the PPT, focusing on the 3A4000.
Loongson’s PPT content is the most solid and thought-provoking. Teacher Hu has developed his own set of theories that are quite persuasive and inspiring.
Loongson’s PPT has almost no flaws; the only flaw that can be pointed out is that the performance indicators of the 3A5000 are somewhat optimistic. According to the PPT, the 3A5000 SPEC06 test results are between 25 and 30 points. This range is very artistic; if it only reaches 25, that is not much of a problem, but reaching 30 might be quite difficult.
Feiteng’s Path: Bright Prospects by Clinging to ARM
Feiteng’s PPT in the first half is about promoting the advantages of the ARM instruction set, believing that the core technology of the X86 instruction set is constrained by others. It asserts that instruction sets like MIPS and Power are not viable, while the ARM ecosystem is flourishing and has the potential to surpass X86.
Feiteng has also brought in Apple, Amazon, and Marvell as external support, believing that ARM will become increasingly popular abroad. They almost claim that ARM is the hope for Chinese CPUs. They then introduced a large number of Feiteng’s applications, below are parts of the PPT.
Feiteng’s PPT has its flaws in that it forcibly elevates ARM to prove itself. The PPT suggests that ARM is mainstream, while other CPUs are on a path to decline.
Some of the wording in the PPT is not very rigorous. For instance, to criticize Loongson, it emphasizes that MIPS processors have weak performance, while the wording for ARM processors is that they have the potential to surpass X86, which is quite artistic. In terms of the critical single-core performance for desktop CPUs, the FT2000/4, using 16nm technology, performs worse than Loongson’s 3A4000, which uses 28nm technology.
Indeed, ARM processors are advancing rapidly and have strong performance, but these ARM processors are from Apple (A13, A14) and ARM Cortex A77. The strength of ARM processors comes from Apple and ARM, not from the strength of FT2000/4.
Using Apple’s and ARM’s processors or cores to argue for one’s bright future is logically unsound and cannot withstand practical scrutiny.
Moreover, the stronger the performance of Apple’s, Amazon’s, and ARM’s ARM CPUs, the weaker the market competitiveness of Feiteng CPUs. Just like in the X86 camp, the stronger Intel and AMD are, the worse VIA fares.
Within the ARM camp, the cost of migrating to a different client is very low. Even if Feiteng has cultivated its market early on, clients can easily switch to competitors’ ARM chips, meaning that having better performance from domestic and foreign ARM manufacturers is not necessarily good for Feiteng. For example, after the release of Kunpeng 920, it has already caused a certain impact on Feiteng in related markets.
Feiteng’s rhetoric is akin to VIA attempting to prove its bright future by referencing the strong performance of Intel and AMD CPUs, which is quite absurd. It only makes one feel that they are playing a game of using foreigners to bolster their own confidence.
Additionally, in this regard, it completely ignores the failed cases of Qualcomm, AMD, and Huaxintong (joint venture, counted as half a foreign entity) and deliberately overlooks the risks of ARM processors, which is also worth discussing.
Zhaoxin’s Path: A Good Cat is One That Catches Mice, Regardless of Color
Zhaoxin starts with a dialectical analysis of openness and autonomy, autonomy and security, which is quite dialectical and has no major flaws.
It then emphasizes the importance of design capability, implying that relying solely on core counts does not represent true capability; strong single-core ability is the real skill. Once single-core performance improves, surpassing those CPUs that merely stack core counts becomes inevitable. This viewpoint is in line with Loongson’s emphasis on single-core performance. AMD has also proven through practice how critical single-core performance is, as evidenced by the 800% increase in AMD’s stock price after the Ryzen series improved single-core performance.
Zhaoxin states that one should not assume that only the red army is autonomous; the blue army is also a vital force against foreign invasion. No matter if it’s a black cat or a white cat, a good cat is one that catches mice. Whether it is a red CPU or a blue CPU, as long as it effectively solves customer problems, it is a good CPU.
Zhaoxin also emphasizes its zero-modification migration advantage, which allows for direct use without the hassle of migration required by other CPUs.
Furthermore, Zhaoxin introduces its GPU, which, although sourced from S3 technology, is not possessed by the other CPU companies. Moreover, Zhaoxin’s GPU is in the first tier domestically, and even if X86 CPUs cannot break ground in the future, it can learn from AMD to engage in differentiated competition with APUs. Below are parts of the PPT.
Zhaoxin’s biggest flaw is also a very interesting metaphor from the conference speaker.
During the conference, Zhaoxin’s spokesperson clarified the X86 CPU technology route and made a metaphor about X86 licensing.
The spokesperson stated that X86 is like a young, beautiful, cultured, and aloof girl. The girl is aloof because she is excellent, so she only opens up to those she finds appealing, and her indifference to the average person does not mean she is not a good girl; she is just too outstanding. In general, cold girls tend to be good girls.
This metaphor is quite effective in whitewashing, making it difficult to refute at first glance. However, according to the analogy of “X86 = aloof beauty,” would not “ARM = passionate beauty” and “RISC-V = available to anyone”?
This metaphor feels somewhat strange.
The key issue is that this metaphor does not fundamentally dispel concerns about X86 licensing, as evidenced by the lessons learned from VIA and Quanta.
Kunpeng’s Path: Spending Huge Sums to Depend on ARM to Create the Strongest ARM-based Processor
Kunpeng’s PPT has the least substance. It starts by discussing the relationship between technology, instruction sets, and microarchitecture, followed by several slides introducing instruction sets, explaining the characteristics of CISC and RISC and their convergence. These are basic knowledge that even an outsider can feel is inappropriate to discuss at such a high-level conference, raising suspicions of filler content.
Kunpeng aims to elaborate on instruction sets, likely to demonstrate that ARM is not inferior to x86. One should not assume that Kunpeng’s ARM CPU is inherently behind x86 CPUs just because it is an ARM CPU.
Next is a discussion of its development history, starting from K3, then 1612, 1616, 920, and 930, claiming that 920 is a generation developed independently. This statement is full of flaws, contradicting the general laws of development.
Then the PPT showcases how impressive Kunpeng is and emphasizes its financial strength.
(Here, a complaint about the term “U.S. dollar”; the Bretton Woods system has collapsed for nearly 50 years, yet the term “U.S. dollar” is still used. If it is not ignorance, it is sycophancy.)
The biggest flaw in Kunpeng CPU PPT is the development history chart, emphasizing that 920 is entirely self-developed.
CPU development is iterative. For example, Loongson and SW. Every generation of Loongson’s CPU evolution and the issues encountered in each generation of CPU development are all made public, making it clear and understandable.
Kunpeng CPUs appear to have evolved from K3 starting in 2009, but in reality, K3 was an integrated Cortex A9, and the CPU core source code was not written by them but purchased from ARM. The subsequent 1612 was an integration of A57, and 1616 was an integration of A72. Moreover, K3 was used in mobile phones, while 1612 and onwards were server CPUs, which are not the same.
From K3 to 1616, the development team merely bought IP for integration, accumulating SoC design experience without independently designing CPU cores from scratch. Even if it is to say technology iteration, it is ARM that has completed the technology iteration from A9 to A72.
To illustrate more specifically, let’s draw an analogy: Lenovo assembles machines from Intel CPUs from the Pentium era to the 10th generation Core. The CPU technology iteration and evolution are entirely done by Intel, and Lenovo gains experience in machine manufacturing, not CPU design experience. ARM CPU companies buy IP for integration, gaining SoC design experience rather than CPU core design experience, as the IP cores are all purchased; ARM is the one iterating and evolving the CPU core.
Interestingly, only Chinese companies have emerged with a batch of CPUs that boast excellent performance while claiming to be independently developed and controllable.
For example, Hongxin, although everyone knows that Hongxin’s technology comes from IBM Power8, does not prevent Hongxin from promoting that CP1 meets independent control.
Similarly, Huaxintong, although everyone knows that Huaxintong’s ARM server CPU technology comes from Qualcomm, does not prevent Huaxintong from promoting that the Shenglong CPU is independently controllable.
Kunpeng has historically relied on buying ARM public versions for integration. Even if we do not count the K3, which was once used in mobile phones and looked down upon, the first ARM CPU (1610) was an integration of A57, the second (1612) was also an integration of A57, and the third (1616) was an integration of A72.
Then suddenly, 920 emerges from nowhere, claiming to be independently developed with 100% self-written source code, akin to a factory with no independent design and manufacturing experience for a MiG-21 claiming it has independently designed and manufactured a J20.
A comparable model is that of American allies introducing F35 technology and spending years learning some of the technology, relying on imported core components while self-manufacturing some parts for assembly, which is a normal technological development pattern.
Indeed, standing on the shoulders of giants is not a problem, but it should not be done hastily. For example, after several generations of iterative research, when significant performance improvements are achieved relative to the original design, the originally introduced source code has been largely replaced, and then it is labeled as independently developed, similar to how Apple developed Cyclone and then promoted it as independent.
Currently, the emergence of domestic CPUs claiming to be independently controllable, with 100% independent research and development, and aggressively entering the institutional market, seeking policies everywhere, is something that does not exist abroad; this has become a characteristic of China.
The Path Dispute is a Struggle for Interests: Seeking Unity of Company Interests and National Interests
The speakers at the conference are all advanced intellectuals, including masters and doctors, and their wording is quite polite. Although there are many implicit criticisms in the PPT, the four CPU companies’ biggest divergence lies in the path dispute.
Loongson firmly follows the independent route,
asserting that the development roadmap should be independent and self-reliant, constructing a system independent of Wintel and AA, achieving a third of the world for itself. This revolutionary development path essentially requires decoupling from the mainstream Western Wintel and AA systems and starting anew. The difficulty is the highest, and if accomplished, it would be of immense significance for information security and industrial development. Once the independent technology system is formed, if North American countries continue to suppress us, we can investigate companies like Intel, Microsoft, and ARM as a countermeasure, allowing them to experience what Google faced back in the day. This is something that other technology routes cannot achieve.
Feiteng and Kunpeng, on the other hand, steadfastly support the ARM route,
with Feiteng’s PPT praising ARM in various ways, not only supporting it themselves but also bringing in Amazon, Apple, and Marvell to bolster their stance, viewing ARM as the future mainstream and hope. Kunpeng dedicates a significant portion to emphasizing that RISC and CISC have converged, arguing that one should not think that ARM, being a RISC, is inherently inferior to X86, and considers itself the strongest ARM-based processor.
Zhaoxin follows the X86 route,
emphasizing that a good cat is one that catches mice, and that the CPU that solves customer problems is a good CPU. Zhaoxin likens X86 to an aloof beauty, suggesting that the stringent licensing of X86 is akin to a beauty being indifferent to ordinary people. The implication is that competitors are envious of Zhaoxin being able to win over the aloof beauty.
The root of this contention lies in the commercial market,
as they all face being dominated by Intel and AMD.
Thus, they can only sustain themselves within the confines of a fenced market.
However, this market is highly influenced by policies, and given the performance and ecology of CPUs, the differences among the four companies are relatively limited and do not reach the level of generational difference. Although one company may have slightly higher or lower data than another, from the perspective of Intel and AMD, they are all just chickens pecking each other, akin to selecting a boxing champion in kindergarten.
Due to the lack of competitiveness in the open commercial market, they can only engage in infighting within the confines of the fence. In the absence of generational differences in product performance and ecology, whoever can secure more discourse power and be recognized by higher-ups as the direction of the “great path” will mean capturing more market share.
Tie Liu believes that fighting for discourse power and market share is not wrong; the key is to maximize company interests while aligning them with national interests.
In the current environment, as North American countries impose increasingly stringent restrictions and suppressions on China, it is imperative to minimize reliance on foreign technology. Whether this task can be successfully completed will determine whether China can withstand the next round of extortion and suppression from North American countries.
If, due to patent licensing issues or foreign companies interrupting licenses or cooperation, it leads to the collapse or shock of a technology system built with enormous state resources, no company can bear such responsibility.
Commercial companies must not allow personal interests to introduce Trojan horses into the secure market, nor should they use the commercial interests of individual companies to hijack decision-making, utilizing vast state resources to build houses on foreign foundations. For those companies that participated in Trump’s ban, such as ARM, a vigilance level of 12 points must be maintained.
Currently, everyone is very cautious about X86, as the PPT states that the share of X86 CPUs in the information technology sector has declined, which is the best proof. However, there is a general lack of vigilance towards ARM, even though ARM has repeatedly complied with Trump’s ban and participated in sanctions against domestic companies.
Now, some manufacturers are packaging ARM as independently controllable, seeking policies and government procurement nationwide; this behavior is akin to squeezing the survival space of independent CPUs, serving as a tool for ARM, and building our independent controllable endeavors on ARM’s foundations, which can be described as causing pain for friends and joy for enemies.
In the current environment, we cannot continue to build houses on foreign foundations, promoting integration into international mainstream and existing foreign systems; this merely replaces one chain of foreign oppression with another. In the ever-changing international landscape,
it is essential to support a truly independent technological system that can decouple from the Wintel and AA systems.
