Who dominates the ups and downs: the battle of the three major routes of domestic CPUs

At 6:00 am on August 10, 2002, with the word “login:” appearing on the computer, there was a burst of cheers in the Institute of Computing Technology of the Chinese Academy of Sciences. The Godson-1 CPU finally worked, and the situation of “lack of cores” in my country’s computers ushered in a breakthrough.

More than a year ago, the Institute of Computer Science of the Chinese Academy of Sciences established the Loongson project and collected 1 million yuan in research and development funds, and 32-year-old Hu Weiwu took over the task. Hu Weiwu was born for Godson. After the college entrance examination, he came to the University of Science and Technology of China in Hefei from the mountainous area of ​​northern Zhejiang. With a living allowance of 20 yuan a month, he spent 8 yuan for “R&D”. And his mentor is Xia Peisu, known as the “Mother of Chinese Computers”.

Godson pinned the hopes of too many people. In order to succeed in research and development, everyone threw themselves in the laboratory day and night, reading Mao Xuan together every day to improve their confidence; moreover, they also gave Godson another name: Gou Left, which carries the meaning of “getting a cheap life to support” in rural China.

Finally, Godson No. 1 was successfully applied, and No. 2 and No. 3 were also successfully developed. At the beginning of 2007, under the leadership of the leaders of China and France, Godson and STMicroelectronics also held a press conference in the Great Hall of the People, and the two sides cooperated to develop the international market. However, a biggest crisis also followed.

In 2009, an article titled “Godson Can’t Buy U.S. Company’s Patent Authorization, the Halo of “China Chip” Is Fading” was published, and Godson’s behavior of spending money to buy MIPS architecture patents was infinitely magnified. For a time, the whole country was in an uproar, thinking that it was developed by itself, but unexpectedly bought it.

At the critical moment, Hu Weiwu had to stand up and give a helpless, sad and red language-style computer principle class for the media teachers who have crossed the threshold. Hu Weiwu explained that buying a chip architecture is not the same as buying a chip; with the architecture, a lot of independent research and development work still needs to be done. However, under the negative shadow of “grinding Hanxin” a few years ago, Godson is still being criticized by thousands of people.

The public opinion dilemma encountered by Godson is also the epitome of the ups and downs of domestic CPUs over the past 20 years. From the difficulty of independence, to the worry of introduction, and then to starting a new business, the domestic CPU has struggled and moved forward despite the ups and downs and disputes. Where does the road to domestic CPUs lead?

  0 1

The first round of autonomy

The first person to propose computer chips was Ni Guangnan.

In 1994, in the face of the massive impact of foreign computers, Lenovo Group, which was advancing by leaps and bounds, was in danger. Ni Guangnan advocates developing in-depth chips, while Liu Chuanzhi wants to build a computer assembly line. “Technology, Industry and Trade” or “Trade, Industry and Technology”? The line battle between Ni Liu and Ni and Liu ended with Ni Guangnan out. At the meeting announcing Ni Guangnan’s dismissal, Liu Chuanzhi took out his handkerchief to wipe his tears, and Ni Guangnan sat in awe, unmoved.[1]

Lenovo’s victory in an era of enlarged trade dividends has affected people’s perception of chips to a considerable extent. Ten years after Ni Guangnan left, Lenovo has published three books: “Why Lenovo”, “Lenovo Storm”, “Lenovo Bureau”. The wide spread of the Ni-Liu dispute, and Lenovo’s 2005 acquisition of IBM’s PC business, seem to be proving a truth over and over again: it’s better to buy chips than to make them, and it’s a fantasy made in China.

When Lenovo became an idol of the times, the domestic CPU that Ni Guangnan fully supported ended in a bleak end.

In order to support Ark Technology in selling domestic CPUs, the Beijing Municipal Government ate the first crab and kicked Microsoft out, which shocked both sides of the Pacific Ocean. The president of Microsoft resigned, Kissinger wrote to the mayor of Beijing to put pressure on it, and finally the prime minister withstood the pressure, and the Ark-1 finally entered the government’s purchase order.[2]

And how glorious the beginning was, how tragic the ending was: no one wanted to use a computer with a domestic CPU, and the boss of Ark Technology was greedy for funds, built a building, and finally said that the money was not enough.[3]This further strengthens the existing stubborn prejudice of the public: domestic CPUs are either thieves or stealing, it is better not to do it, or Intel is the most fragrant.

However, unlike the perception of consumers, the computer veterans of the Chinese Academy of Sciences continue to appeal to the country: to start the research and development of independent CPUs during the “Tenth Five-Year Plan” period, otherwise there will be no chance. Finally, dissatisfied with Intel’s monopoly position and considering the security risks brought about by CPUs being controlled by others, the country has introduced the “Taishan Plan” aimed at revitalizing domestic independent CPUs.

At the same time as the Ark, which had high expectations at the time, there were three other domestic CPU projects:

  GodsonDue to the lack of funds, even the cheap MIPS architecture could not afford it at the beginning;

  Shenweideveloped by the 56 Institute with a long history in Wuxi, using the ancient Alpha architecture to serve the military and supercomputing;

  soardeveloped the first CPU as early as 1999, using the SPARC architecture.

During the “Tenth Five-Year Plan” and “Eleventh Five-Year Plan” period, with the support of many state-level funds such as 863 and nuclear high base, domestic CPU chips ushered in the spring breeze: basically maintained the production of one every two years. The speed of the chip, and spare no effort to catch up with the performance. The Godson II in 2004 can already equal the Pentium III in 1999.

However, the three major domestic CPUs also quickly encountered three major difficulties:

  performance difficultiesIntel’s combat effectiveness has exploded after entering the new century, AMD was disabled in 2006, and domestic CPUs have made great progress, but there is always a generation gap;

  ecological difficultiesthe x86 patent wall is towering, and the domestic independent CPU adopts a detour strategy, but it is difficult to adapt to the Microsoft system;

  Mass production difficultiesusers have adapted to Microsoft’s operating system, a barren desktop, unattractive, how to talk about mass production.

The reason for these three difficulties is that domestic CPUs have adopted a reduced instruction set architecture: MIPS, Alpha, SPARC, etc. These architectures that are unheard of by ordinary people are all defeated by Intel. The core element is the alliance formed between Intel and Microsoft.

With the end of the “Eleventh Five-Year Plan” and the tightening of state subsidies, the three major domestic CPUs have ushered in a turning point in their fate: Hu Weiwu gritted his teeth and decided to let Godson take the road of marketization, and the R&D personnel lost their iron rice bowls and went to sea; Shen Wei hid in the system and the supercomputing market, and did not attack downwards; a dispute broke out in Feiteng, and the two factions were arguing in front of the government leaders whether to abandon the original structure and fell into confusion for a while.

Can’t see the dawn of domestic CPU, so someone proposed another route: since it can’t be beat, why not introduce it?

 0 2

The second round of introduction worries

While large planes, nuclear Power plants and high-speed railways are beginning to become autonomous, the introduction of CPUs has begun.

On the introduction route, two armies were dispatched in China: Shanghai Zhaoxin and Tianjin Haiguang. The most popular and commercialized chip in the industry is Intel’s X86 architecture chip, but Intel is very strong and difficult to negotiate. Therefore, both companies chose to “save the country by the curve”:

Shanghai Zhaoxin acquired the X86 license of VIA Electronics, a company in Taiwan. VIA is known as “Taiwan’s Intel”, and its X86 patent license comes from the two American companies Cyrix and Centaur that it acquired that were on the verge of being beaten by Intel. VIA Electronics-related companies are also the founding shareholders of Shanghai Zhaoxin. Subsequently, Zhaoxin also acquired the X86 patent of VIA Electronics for more than 200 million US dollars.

Tianjin Haiguang chose AMD. AMD has obtained a cross-patent authorization for Intel X86 very early, and its self-developed Zen architecture has powerful performance. However, due to the fierce competition from Intel, AMD’s cash flow is under great pressure, and blood transfusion is urgently needed. Therefore, there is a cooperation with Haiguang, but AMD is also careful: Haiguang needs to be authorized by two-tier companies and go through 11 processes. Products are sold in mainland China.

After the introduction of technology, Zhaoxin and Haiguang have greatly increased their strength. Zhaoxin’s deal with VIA can be described as killing three birds with one stone, obtaining a CPU R&D team, x86 architecture patents, and a chipset industry chain; while Haiguang relies on the listed company Sugon, drawing on AMD’s work, plus domestic encryption algorithms, in commercial use The server starts quickly.

However, whether it is buying or renting, the price is high. Zhaoxin and Haiguang spent more than 2 billion yuan each; but it did make a sound very quickly. Two years after its establishment, Zhaoxin achieved mass production and sold more than 10,000 sets. Haiguang’s revenue in the first half of this year was 270 million and net profit was more than 60 million.

The introduction route seems to be immediate, but when Haiguang was included in the U.S. entity list, AMD, who was relieved, said that the latest architecture would no longer authorize Haiguang; and Shanghai Zhaoxin also faced patent flaws, we discovered that the introduction is not a panacea.

At the same time, the x86 talents in mainland China are extremely weak, and there is a big question mark as to whether the “bringing doctrine” can evolve into “absorbing innovation”. If you can only stay at the copy and paste of the x86 hardware level, it is difficult to carry out in-depth research and development, and meet the needs of high security level and consumer market.

Independence needs to start from the bottom, which is almost impossible; the introduction of the X86 architecture is also faced with the worry of being cut off and unable to update. When the market was tangled again, ARM came into view.

 0 3

The third round starts all over again

Since entering China in 2001, ARM has given the impression that it is an “old friend of Chinese technology companies”: reliable, easy to use, and capable of playing. More importantly, the ARM architecture authorization can be permanently bought out at the instruction set level, which is more controllable on the one hand, and deeper in research and development on the other hand. Therefore, domestic CPUs began to be independently developed based on the ARM architecture, trying to achieve the same performance as the x86 architecture, so as to achieve a deeper substitution.

Facts have proved that domestic CPUs are doing really well, and they are assisting ARM.

After going through the “autonomous VS introduction” line debate in the fall of 2011, Feiteng chose the ARM architecture. In August 2015, a Chinese named Charles Zhang (Zhang Chengyi) introduced the Feiteng server CPU “Mars” in Chinese on HotChips, the top conference in the global chip industry. The world’s first ARM-based CPU The 64-core CPU developed by the architecture.

This release caused quite a stir in the industry, but then the domestic public opinion made Feiteng experience the cruelty of Vanity Fair: Feiteng’s “comparable to Intel” propaganda was “shocking”, and the public thought it was an inexplicable manuscript reading style The press conference must hide the shocking secret of deceiving state subsidies. Some people even called on the masses to launch a monitoring campaign against domestic CPUs on the Internet.

Feiteng fell into a Schrödinger public opinion field: if you don’t do it well, some people will criticize it, but if you do it well, there will be people who will criticize it. And this is almost the situation faced by many domestic CPU companies. To break through the predicament, only consolidate strength:

Feiteng’s commercial server and desktop CPU processes have reached 28nm and 16nm respectively, which are basically suitable for use in specific domestic scenarios; and combined with the Kirin operating system developed based on open source Linux, 2 million Android applications can be run on desktop computers; at the same time, through capital acceleration , Feiteng is integrated into the listed company China Great Wall, which is convenient for supporting the whole machine.

The rapid awakening of domestic substitution awareness in the past two years has put Feiteng on the fast track. In 2018, Feiteng sold 200,000 CPUs based on the ARM architecture. In 2019, its revenue tripled. In the first half of 2020, its revenue was 340 million, which has exceeded the whole year of last year. And its leap-forward development, thanks to the multi-channel, “rural encircling the city” strategy, products cover embedded, desktop and traditional servers.

Just when Feiteng expanded with the help of the ARM architecture, Huawei got involved in the battlefield, and the first shot was a big army battle, and it was high in the server CPU market.

At the beginning of 2019, Huawei released the first ARM-based server CPU “Kunpeng”, and established the Kunpeng Computing Industry Alliance with various provinces and cities to deliver industry standards and talents to various places, and established hardware production bases with representative enterprises in various places. In May of this year, Kunpeng CPU entered China Telecom’s procurement list.

Huawei’s aggressive offensive, just like the way it played in the program-controlled switch market, was bound to local telecommunications bureaus and co-existed. Finally, with technical upgrades and channels, the localization of switches was completed.

The domestic CPU gave up Intel and started a new ARM stove, and it was indeed very prosperous. However, ARM is also swaying under the US ban, and there is even a risk of being sold. Therefore, some observers believe that the name of the domestic CPU based on the ARM architecture is fake, and it is ARM. Independent and controllable: If you cut off your confession, you will also suffer. With an outdated permanent structure, you will be defeated by others sooner or later.

After going around in circles, the question returned to the original point: what degree of autonomy do we want to achieve? What steps should be followed for localization?

 0 4

epilogue

From 2000 to the present, domestic CPUs have followed three routes:

The independent faction is obsessed with fighting against Intel’s x86 architecture, taking the defeated original architectures such as MIPS and Alpha as their own, starting from the most basic instruction set, and constructing a CPU castle that the Chinese enjoy full control:

The market faction has continued the consistent technical route since the reform and opening up: introduction, digestion, absorption, and re-innovation. Due to Intel’s patent wall, they have to adopt a roundabout strategy, or buy or rent the castrated x86 architecture. The road to digestion is still not yet there. know;

Eco-style, relying on its own R&D strength, tries to leverage its strengths in the competition between ARM and Intel, taking advantage of the newly emerging ARM ecosystem to gradually erode the existing x86 market.

The three seemingly different routes are actually for two major demands: autonomy, safety and controllability, and meeting market needs. However, these two demands are contradictory to a certain extent.

Fully autonomous and controllable means reconstruction, screening and demining at the instruction set level. In a strict sense, X86, which is the most closed but monopolized, should be the first to be excluded. However, x86 is the most successful commercialization. Users are completely accustomed to the combination of Microsoft + Intel. The migration cost is the lowest along the x86 route.

This is the embarrassment of domestic autonomy. Different from the independent and closed production of atomic bombs, CPUs, operating systems, and chips are all products with strong commercial attributes, and the industrial chain and development are quite international. ideology” is constantly balancing.

In the past, it was easy for us to fall into the trend of “making is worse than buying,” but today’s external pressure reminds us that although autonomy is difficult, exploration cannot stop.

The Links:   M170EG01-VH GP470-EG11