[Reprint] China simmering core, Godson / soar in what way to go

China simmering core, Godson / soar in what way to go

Blue Sky · 2015-10-23 · local IC  sources:  cnBeta
  In recent years, support for nuclear high base project subsidies and national IC industry investment fund, the number of domestic engaged in the design of high-performance CPU units or growing companies, among which there are like Godson, soar, Shen Wei, which has had a profound technology IC design heritage of the old units, but also as macro core, core trillion this rookie; both Spreadtrum such a state-owned holding company, also has such a non-state-owned enterprises Hass.
  Some people commented it was "television flourishing", it was evaluated that is "redundant construction, and mutual strife."
  In fact, in dazzling design units and companies, according to the level of self-control and the degree of market-oriented operation of difficulty, it can be divided into three difficulty modes:
  STRIKE independent route, Hard mode technology to build its own system, which represents Godson, Shen Wei;
  the other is to design their own micro-structure, ensure a safe and controlled chip, but dependent on Wintel or AA system, its software is compatible with ecological Normal mode, represented soar, the king is, Chung Chi;
  the last one species are foreign manufacturers and mainland cooperation, joint venture, or completely dependent on Simple mode AA system software and hardware, on behalf of the former is the core trillion, on behalf of the latter is the core macro Hass, Spreadtrum.
  Here, we count domestic CPU technology roadmap and market prospects for the development of the route from three difficulty modes.
A, Hard mode roadmap
  independent intellectual property rights in the name suggests roadmap, roadmap choice is totally in charge of their own, to the independent route has the following characteristics:
  First, with independent right to development.
  Has its own set of instructions, can independently extended instruction set, you can choose in the direction of development. For example Godson At the same time obtain permanent authorization of Mips, self-expanded 148 loongEXT, 5 Tiao loongVM instruction, 213 loongBT, 1014 Tiao loongSIMD, the original 527 Mips instruction, extended to 1907, developed into Godson own loongISA Shen Wei of the Alpha instruction set also a large degree of expansion.
  The second is a safe and controlled.
  To build their own technology system can be implemented to achieve a safe and controlled from both software and hardware. For example, Godson is to "loongISA + LCC + GS464E and other self-developed micro-structure + Community operating system loongnix + software ecosystem + Industry Alliance" as the basis, and strive to build their own system, due to software and hardware are built himself, and therefore high security . Shen Wei is also actively build their own software and hardware systems, has developed its own divinity wisdom compiler, developed a Linux-based operating system divinity Koch, coupled with ultra-count areas like ecological constraints exist as PC software field, Shen Wei Construction of the road independent technology system can be described as a smooth sailing - in September 2012 and put into use the supercomputer divinity Blu-ray uses a 8704 Shen Wei 1600, equipped with divinity Koch operating system, software and hardware all domestic. (Godson GS464E line schematic)   Third, all the profits in the country.   For example, Godson built a covered twenty-three hundreds of enterprises, tens of thousands of industry alliance technicians. Because with independent intellectual property rights, we can achieve all profits remain in the building of Godson Industry Alliance, foreign firms unless industry alliance to join the Godson, Godson when the messenger to, or can not take a single cent of profit points. Shen Wei due to focus on supercomputing field, industry alliances relatively closed, liquid cooling system in addition to the current supercomputer divinity from outside the foreign buyers, most of the rest to achieve localization.   There are also two disadvantages under Hard mode:   First, the technical threshold is high, industrialization difficult.






  To complete the design unit independent instruction set extensions, micro-structure design, compiler development, ecological construction and other software work, but also to attract more than enough, strong enough to join the partnership, the establishment of industry alliances.
  The second is not conducive to market-oriented operation.
  Currently, PC field dominated by Wintel systems, and mobile chip is the end of the world AA system, the basic low-end market occupied X86 server chips aspect, and high-end servers is IBM's traditional markets. Build your own system necessarily incompatible foreign Wintel systems and AA system is not conducive to market-oriented operation.
Two, Normal mode roadmap
  Normal mode roadmap refers to the purchase of foreign instruction set design units authorized, independent design microstructures, but dependent on Wintel or AA system, its software compatible roadmap ecology. Typical: FT (ARM), the king is, Chung Chi.
  Normal mode has the following two characteristics:
  First, the purchase of foreign instruction set authorization.
  For example, soar purchased ARM V8 instruction set authorization, the king is authorized to buy MIPS, X86 PKUnity is obtained authorization (Godson discussion group set up, AMD's initiative to come to. As for the reasons, you know) . (National Defense Science design compatible ARM V8 instruction "millet")   Second, independent design microstructures and CPU, up to protect the hardware from the hardware security control.



  For example, soar designed the "millet", and "Mars" and "Earth" two products; the king is designed XBurst0, XBurst1, XBurst2, and Z-series, X series, M series three product lines; north Volkswagen Chi independently designed UniCore-1, UniCore-2, and Jushi -805, PKUnity-3-130, PKUnity- 3 (65), PKUnity86-2, PKUnity86-3 CPU or SOC, which has a very high PKUnity86-3 integration, to achieve the integration of CPU, GPU, DSP, Northbridge and Southbridge chipset into a single chip, the greatest extent possible to circumvent the foreign companion chip stock large quantities of machine production and downtime risk. Third, foreign Wintel eco-compatible software system, AA system.   For example, "Mars" and "Earth" can run smoothly Android and its applications; it is because the king is the large number of applications software is written for ARM, leading the king is unable to run these applications, address this issue, the king is being MC transcoding developed, to improve compatibility problems in some extent. In addition, the king is equipped with a smart watch inWatch T TencentOS; PKUnity is fully compatible with Window98, WindowXP, Window7 other operating systems. (Jun is used for smart watch inWatch T of Ingenic Newton2 - wearable devices open source development platform)   While in Normal mode, the unit can not be self-designed instruction set extensions, only picked up the pace of technological development in accordance with the road map delineated foreign, can not independently choose their development direction. And whether it is authorized to update the expiration or instruction set, instruction set should be authorized to buy again, is always controlled by others, became senior employees.   But in the absence of their own system or can not rely on self-built system of capital and technology, the system dependent on foreign countries, is the only alternative to compromise the independent development of the market-oriented operation.   The pros and cons accompanied by a mixed blessing. Normal mode also brings the following advantages:   significantly reduced First developed technical threshold, time and capital costs.
 








  For example soar and the king is not need to own instruction set extensions, the compiler does not need to develop their own, but do not need to build software ecosystem and Industry Alliance, only we need to design "millet", XBurst and UniCore-2 and so on microstructure and "Mars" , M150, M200, X1000, PKUnity86-3 can. (Jun positive things for the M150)   Second, it helps the market operate.   After attached to the foreign system, compatible with its development has been very mature and feature-rich software ecosystem, will not face bottlenecks in the software ecosystem Hard mode. This soar, and the king is Zhongzhi market operations is a big plus.   Three, Simple mode roadmap   Simple mode there are two cases:   one is the mainland and foreign manufacturers of cooperation, joint venture;   the other is completely dependent on the software and hardware aspects of AA system.   The former representative trillion is the core, the core macro, which is represented by Hass, Spreadtrum.   For trillion as the core, the core macro and foreign manufacturers such joint ventures or cooperative enterprises, the advantages are obvious, shade trees are good - Mega core can be sheltered Wintel systems, while IBM Terawins able to get technical support. Of course, the cost is also huge. From now on, trillion-core becoming cowboys Wintel system, the macro core IBM is also tied to the chariot OPENPOWER, always lost like Godson possibility of self-development as Shen Wei.   Moreover, the introduction of technology, often have such a process, the first OEM, after imitation, and then modify the original design, the final innovation after the introduction of the technology in context. Therefore, Terawins released this year CP1, in fact, IBM's Power8 vest, the only core trillion a desktop chip ZX-A vest is the VIA Nano.












  According to PPT trillion core and core macro published, to finish the OEM, reproduce, modify, innovation fundamental to this process to around the 2018 ZX-E and CP3. Before this product, basically not to mention intellectual property rights, let alone safety control.
In addition, mega and macro core for core, there are other risks:
  For Terawins, IBM technical support for macro core is restricted POWER 8 of the most valuable aspects of the floating point arithmetic technology is not Terawins! Open. In addition, IBM will transfer technology to the macro core intention is to support a wage earners, do quad-core or eight-core processors and Intel to grab the market.
  If you want to use the macro core IBM technologies to design their own micro-structure, do high-end server CPU and IBM compete, then the macro will not be the core of IBM's technical support.
  For trillion core, the risk is even greater.
  Taiwan's VIA X86 patent from the acquisition of Cyrix Corporation, although X86 won the right to use part of the patent, but the lack of VIA heritage in terms of technology accumulation, in the patent stick attack suffered Intel, VIA's X86 chip disappeared.
  Although Intel and the Federal Trade Commission (FTC) to reach an agreement, Intel must provide the x86 licensing agreement was extended to April 2018 to VIA - in April 2018 ago, VIA and Intel will be able to use each other's patents, but after April 2018, VIA can not use Intel's new patent (patent can continue to use the old).
  So, to 2018, the core if not trillion Intel's new patented technology, it is likely to be far more in terms of performance Intel widening, if the use of Intel's new patent, then it is likely to be a big wreck Intel patents hemiplegia, repeat the tragedy VIA year.
  In addition, the core technology starting from a low trillion. Cyrix there are a large number of technical backbone of the company as early as brain drain in the year 2000, VIA's X86 chip in around 2009 on the basic technological development stalled, VIA official website is not updated million years, the best product is the VIA Nano, but its performance is really miserable head forbearance see, do not say and Intel haswel comparison, and is compared Godson GS464E have a big gap. (Chart for the SPEC2000 test scores at 1G frequency, CPU unspecified sections of the build environment, for reference)   , but the latest VIA QuadCore E C4650 has some progress, in terms of process technology and frequency than the Nano has some improvement, but according to official website data, the microstructure is still small change version of Isaiah, single-threaded capability is still relatively weak.   In addition, do not know whether to transfer the VIA QuadCore E C4650 trillion to the core, QuadCore E C4650 is whether trillion core ZX-C. For like Hass, companies such as Spreadtrum   process dependent on AA system under conducive to business operations, earn quick money, but buying ARM micro-architecture, integrated in its own SOC, ARM microstructure license fee and per production an ARM chip like royalty tax vampire, squeeze the IC design company profits.   Hass and Spreadtrum life's very moist, because Huawei Hass is a subsidiary of Kirin chip homegrown worry about selling, pricing on the market is not affected, Huawei inside the final say, and the high premium Huawei's products can Hass is to allow sufficient profit margins.   Spreadtrum acquired after being purple, Intel Youyi $ 1.5 billion stake in Spreadtrum get 2 percent stake in the fund to give other countries a large purple IC 40 billion. In purple and Intel transfusion, Spreadtrum is naturally capable of maneuvers. And those who can not "fight father", or although there is a father, but the father is not Gang of ARM camp vendors, such as LCI, Actions, new shoreline, Chi, Rockchip and other not so lucky.








  In addition, because they are buying ARM micro-architecture, bringing a high degree of product homogeneity, lack of core technology, profits generally low, fierce market competition, the survival pressure, there is the risk of product safety and other issues. Fourth, choose a direction, focus on tackling feasible?   Currently, if you count was terminated soar developed SPARC, in the country can be described as assembly of the X86, ARM, MIPS, Alhpa, POWER, SPARC six camps, but also each engage a sets of software ecosystem and industry Alliance, said that the repeated construction, mutual strife are not unreasonable.   So, the state is able to mediate and choose 1-2 main direction to achieve focus, focus on tackling it?   Hard!   In addition to sectoral interests is difficult to reconcile, the routes do not agree with each other is also an important factor in the development - in the Godson, Shen Viagra appears, select Normal mode, Simple mode, always controlled by others, can not form their own software ecosystem can not choose the direction of development, can never achieve independence.   For Godson and Shen Wei, one has the strongest micro-structure, and the other has the best high-performance many-core CPU, after ten years of accumulated technology, Godson and Shen Wei already have their own instruction set, compiler , a prototype of the software and considerable ecological and industry Alliance, to give up their existing technology accumulation, switch to X86, ARM camp is obviously impossible.   In trillion core, Hass et opinion, Hard mode only select high technical barriers, market-oriented operation even more difficult on the blue sky, the United States is fighting armed Tuba Lu division, the kiss of death. When the shade of the Wintel system can be attached to the system of large trees and AA impossible to select Hard mode.   In addition, Hard mode is successful is good, whether it is on information security or commercial interests are second to none, but the risk is unusually large, failure is down the drain. The Normal mode, Simple mode, although gains in information security or commercial interests and Hard mode can not be compared, but with lower risk, in the case of the West behind the overall domestic semiconductor industry, the only alternative to the reality of compromise.









  Therefore, the state must also decision-making difficult choices, then to a "Eight Immortals recount," to be able to go all the way to go again, there is always to be able to. But the Chinese people and the accumulation of capital investment in IC design, we really have the time and capital to squander it?
Fifth, why the state of various IC design companies to support the huge differences
  inventory of each company to obtain state funds to support the situation:
  2014 in violet (Holdings Spreadtrum) integrated circuits large national fund 40 billion yuan.
  Won the macro core set up registered capital of 600 million, from its inception in 2014, obtained grants of not less than 2 billion yuan.
  The signs core set up the Shanghai municipal government invested 1.2 billion yuan, from 2013 to now set up to obtain subsidies not less than 70 billion yuan.
  Godson Since its inception in 2001, from 863 countries, nuclear high base accumulated to obtain special project funding 500 million yuan. Godson in the Division of the company, the Beijing Municipal Government equity investment of 200 million yuan. No access to national funds to support the project after 2010.
  Soar, Shen Wei is a result of military projects, and there is no data.
  Under China's national conditions, access to technology does not rely on state subsidies of various technological achievements, technical route (technical level the best and most abundant technological achievements Godson, Shen Wei, soar, obviously not as good omen core, Spreadtrum by the state pay attention), but on the ability to influence public relations and the international giants of the various units in China.
VI Conclusion
  five camps strife with each other, one is duplication and waste of resources, but also brought intense competition, I believe in the fierce market competition and survival of the fittest, Chinese core can rise from the ashes, a place in the international chip market.
  Note: This article reprinted from other media aimed at passing reproduced more information, does not mean Ben Wang agree with their views and be responsible for its authenticity.

Guess you like

Origin www.cnblogs.com/jinanxiaolaohu/p/11477062.html