"Compared with godson, which has been at the forefront of public opinion since its birth, Shenwei, another chip in China that takes the road of full autonomy, is much lower key. Whether it is traditional paper media or online media, Shenwei's exposure rate is much lower than that of domestic chips such as godson, Hisilicon and Spreadtrum. Moreover, compared with Godson with full talent and skills, Shenwei rarely becomes the target of attack by online media and abusive Internet users.
Shen Wei's military background gives it a hint of mystery. Next, let's walk into Shenwei and decrypt the "core" process of Shenwei.
1. Birth of Shen Wei
From the 1960s to the late 1970s, China also had its own glory in the field of high-performance computers - Harbin military industry once manufactured 441b series computers with time-sharing operating system, assembly language, FORTRAN language and standard program library. Peking University, Beijing cable power plant and other units jointly developed 150 computers. Tsinghua University and Beijing Radio No. 3 factory have developed and produced nearly one thousand 130, 131, 132, 135, 140, 152 and 153 series computers. The University of national defense science and technology has successfully developed 151 computers, and the East China Institute of computing technology has successfully developed 1001 medium-sized integrated circuit computers and HDS-9 computers. Among them, HDS-9 computers operate 5 million times per second. What is more valuable is that the software and hardware of these computers are compiled and manufactured by the Chinese themselves.
However, in the 1980s, because of superstitious theories such as "buying is better than making, renting is better than buying", "market for technology", they gave up the research and development of independent chips and purchased a large number of foreign chips. From galaxy 1 supercomputing in the 1980s to Galaxy 2 and Galaxy 3 in the 1990s, dawn series supercomputers have no Chinese "core" and are always controlled by others in technology. Even on Tianhe 2 a few years ago, there were only 4096 Feiteng 1500 as the front-end processor of the computing node, and the Feiteng 1500 is an improved version based on sun's UltraSparc T2.
In order to solve the dilemma that there is no "core" available in supercomputing, national defense and information security. The 56th Research Institute of the general staff (Wuxi Jiangnan Computing Technology Research Institute) began to design Chinese people's own high-performance chips in 2003. Shen Wei embarked on the journey.
(super cloud Shenwei server)
2. Shen Wei's family
Shen Wei is designed by the 56th Research Institute of the general staff (Wuxi Jiangnan Computing Technology Research Institute), and Shanghai high performance integrated circuit design center is the vest of the 56th Institute.
The Institute was founded in June 1951 and is located in Wuxi. The main research interests include computer system structure, parallel / distributed processing and intelligent computer system, computer software, artificial intelligence and intelligent control, computer and communication, computer information processing and application. Since the 1980s, it has won more than 300 national and military scientific and technological achievement awards in the fields of computer, communication and mechatronics, including 3 national special awards for scientific and technological progress and 35 first prizes for military scientific and technological progress. In 1987 and 2000, the Central Military Commission twice won the collective first-class merit for 56 institutes. In 1993, the Central Military Commission awarded 56 institutes the honorary title of "Advanced Research Institute for bravely climbing the peak of science and technology".
(thin client)
3. Instruction set and technical origin of Shenwei
Shenwei belongs to the alpha camp, and the instruction set is also extended based on alpha.
Shenwei's technology comes from Dec's alpha 21164. Dec's technical strength is very strong, and alpha's performance is even more amazing. Amd once purchased the technical data of alpha 21264 and developed its own K7 microstructure after poaching some Dec technicians.
Dec is a typical example of technical strength stronger than people, but the business model is not as good as people. Alpha has also been changed hands several times, first acquired by Compaq and then hidden by HP. At present, alpha has been shelved, the instruction set and microstructure are no longer updated, and most technical patents have expired or are about to expire. Shenwei is the only remaining fruit in the alpha camp and has the autonomy to expand instructions and development routes.
As for the purchase of foreign instruction sets and the expansion of instructions based on them, and whether the development of their own instruction sets is autonomous and controllable, please refer to another article I wrote, "what does Godson take against foreign chip giants in 15 years".
(firewall)
4. Construction of autonomous and controllable system
Shenwei and godson, as the only two IC design units taking the independent and controllable route in China, are actively building their own software and hardware system like Godson to ensure the safety and controllability of software and hardware.
Shenwei has developed its own Shenwei Ruisi compiler and developed Shenwei Ruisi operating system based on Linux. In addition, the supercomputing field does not have the constraints of software ecology like the PC field, so Shenwei's road to building an independent technology system can be described as a smooth road - Shenwei Blu ray supercomputing, which was put into use in September 2012, used 8704 Shenwei 1600, equipped with Shenwei Ruisi operating system, and realized all localization of software and hardware.
5. Shen Wei process
56 Institute (Jiangnan Institute) expanded its own instruction set based on alpha instruction set, and designed Shenwei 1 with independent microstructure in 2006. Shenwei 1 is a single core CPU with 130nm process technology, with main frequency of 900MHz and integrated 57 million transistors. Shenwei 2, completed in 2008, is a dual core CPU with 130nm process, with a main frequency of 1.4GHz.
If Shenwei 1 and Shenwei 2 are more like the works honed by 56 and the digestion and absorption of alpha21164 technology, Shenwei 1600 developed in 2010 is 56's new products after integrating foreign technologies. Shenwei 1600 is a 16 core CPU with 65nm process technology, with main frequency of 1.1g and double precision floating-point 140g. Shenwei 1600 is used for Shenwei Blu ray supercomputing.
Perhaps because of the strong performance of alpha in those years, I have high expectations for Shenwei from alpha. However, according to the paper published by the national high performance integrated circuit design center, the spec2000 test score of Shenwei 1600 can only be said to be unsatisfactory, and only the performance power consumption ratio and Linpack test can be comforted.
To be fair, although the single core performance of Shenwei 1600 processor is limited, the scalability of 16 cores is very good, realizing an ideal multi-core performance acceleration ratio.
The performance parameters of Shenwei 1600 are shown in the figure below:
In 2012, 56 institutes developed Shenwei 1610 and Shenwei 410. Shenwei 1610 is a 16 core CPU with 40nm process, integrated 1 billion transistors, main frequency of 1.6g, maximum power consumption of 50W and double precision floating-point operation of 200g. Shenwei 410 is a 40nm process 4-core CPU with 270 million transistors and a main frequency of 1.6g. The former is used for server and the latter is used for PC, which supports the winning Kirin operating system. These two products also encounter the dilemma of lack of market competitiveness caused by poor software ecology and weak industrial alliance encountered by Godson in the PC market. In addition, the single core performance of Shenwei 410 is too weak, and the safe and reliable PC equipped with Shenwei 410 basically does not have the competitiveness in the civil PC market.
By the end of 2014, Shenwei 5 was successfully released. According to a paper jointly released by the State Key Laboratory of digital engineering and advanced technology in Wuxi and the national parallel computer engineering technology research center in Beijing, Shenwei 5 is likely to be a high-performance multi-core CPU integrating 4 management cores and 256 operation cores, with double precision floating-point operation of more than 1t and in core Linpack efficiency of 93%, It has high performance power consumption ratio.
(structure drawing of Shenwei 5)
6. Shenwei Blu ray supercomputing
Shenwei Blu ray supercomputing, with a total investment of 600 million yuan, was put into use in September 2012. Shenwei Blu ray supercomputer uses 8704 Shenwei 1600 and is equipped with Shenwei Ruisi operating system to realize the localization of software and hardware. The peak computing performance of Shenwei Blu ray supercomputing is 1pflops (10 billion times per second), the continuous performance is 0.796pflops, the performance power consumption ratio exceeds 741mflops / w (million floating-point operations / second • Watt), and the Linpack efficiency is 74%.
Shenwei Blu ray supercomputing has the following characteristics:
All use Shenwei 1600 chip and Shenwei Ruisi operating system to realize the localization of software and hardware;
Good stability, no fault in the whole process of Linpack test for more than 9 hours;
The liquid cooling system is efficient and reliable. The closed water circulation of the coolant inside the cold plate takes away the heat of the main plate, which is environmentally friendly and noise free;
High density packaging, 1024 CPUs can be installed in one machine bin;
The whole machine has low power consumption and high performance power consumption ratio. The performance power consumption ratio of Shenwei Blu ray supercomputer exceeds 741mflops / W, and the power consumption of the whole machine is 1W; The performance power consumption ratio of Tianhe 1 supercomputer is 431.7 MFLOPS / W, and the power consumption of the whole machine is 4W.
High efficiency of the whole machine. The efficiency of Shenwei Blu ray supercomputing Linpack under different scales: single processor core 80.28%, single CPU 75.20%, single cabin 75.07% and whole machine 74.37%. The efficiency of Shenwei Blu ray supercomputer is equivalent to that of Jaguar supercomputer, which was once the fastest computer in the United States.
According to statistics, the "National Supercomputing Jinan center" has more than 100 registered users and more than 60 key application topics for transplantation and optimization, including 16 application topics that can reach the parallel scale of tens of thousands of cores, more than 15 topics with application results, and the system utilization rate has exceeded 60%. At present, it mainly faces the application fields of Meteorology and climate, marine environment, biomedicine, information security, aerospace, material physics, financial analysis, industrial design, petroleum geophysical exploration and so on.
New generation Shenwei supercomputing
In 2013, Wuxi Municipal People's government established a National Supercomputing Wuxi center construction leading group with the mayor as the leader and the vice mayor as the deputy leader to actively prepare for the construction of a new generation of Shenwei supercomputing. In April 2015, Wuxi Hengding Supercomputing Center Co., Ltd. was officially established as the main body of the construction and operation of the National Supercomputing (Wuxi) center, and 56 Institute established a project command team and chief engineer team to solve the technical problems in the process of supercomputing construction.
At present, the new generation of Shenwei supercomputing host system software is being debugged, and the whole machine system design is planned to be announced at the end of 2015. The total investment of new supercomputing is 1.8 billion yuan. It is planned to be completed in 2017-2018. The computing capacity will reach 100pflops, and the computing performance is expected to hit the first in the world.
Shenwei MPP machine
Shenwei MPP machine is a high-performance computer with a scale of more than 10 billion developed by the national parallel computer engineering technology research center. It can virtualize a variety of computing and storage service forms in the service layer through virtualization technology, grid computing technology and cloud storage technology.
The hardware system of Shenwei MPP consists of high-speed computing system, interconnection network system, mass storage system, maintenance and monitoring system, power supply system, cooling system and structural assembly design. The software system mainly includes "Shenwei Ruisi" operating system, "Shenwei Ruizhi" compiler, basic library, etc.
Since the system was put into use, it has successfully transplanted nearly 200000 parallel application projects in key application fields such as climate and meteorology, marine science, new drug development, biological information, aerospace, industrial design and financial analysis, and achieved several important application achievements.
6 Shenwei cluster is a high-performance computing platform specially designed and developed for the field of scientific and Engineering Computing Based on the technical advantages of Shenwei MPP machine. It is widely used in climate and meteorology, marine environment, numerical wind tunnel, collision simulation, protein folding, gene research, new drug research and development, molecular dynamics simulation, quantum chemical calculation, material science, chip design, animation design, industrial design, etc.
The general structure of Shenwei cluster system is shown in the figure below:
7. Secure and trusted PC
Shenwei secure and trusted PC is equipped with Shenwei 410. For party, government and military units with high security needs, the components are 100% localized. The embedded domestic peripheral chip integrates the trusted password module in line with the national standard to realize the embedded trusted guidance of the system and the trusted measurement of the software system, so as to ensure the security and credibility of the system.
Because the single core performance of Shenwei 410 is too weak and the software ecology is extremely poor, it is limited to meeting the needs of special departments. If Godson is still developing from the security market to the civil market, Shenwei has never considered the civil market from beginning to end. It is completely closed and played by itself.
Realistically speaking, the safe and reliable PC equipped with Shenwei 410 is not only expensive, the overall performance is weak, but also the software ecology is poor, so it basically does not have the competitiveness in the civil PC market.
After Shenwei 5 took the public nuclear route, since 2012, Shenwei has no new desktop chip products and is likely to have given up desktop chip development.
The security architecture is shown in the figure:
8. Shenwei cloud
Shenwei cloud is a cloud product developed by the national parallel computer engineering technology research center. It can provide rich cloud product suite, open API interface, comprehensive resource management platform and operation and maintenance monitoring platform, and has strong virtual resource pool management ability. Shenwei cloud platform adopts commercial servers, high-speed networks and high-performance storage devices, combined with Shenwei cloud platform series software. It has the characteristics of powerful function, stable performance, safety and reliability, simple and easy to use.
9. Summary
Although Shenwei and Godson follow the independent route, Shenwei's road is much smoother than Godson:
First of all, the state's investment in Godson is very limited, and Godson has fully operated independently and is responsible for its own profits and losses since its market-oriented operation in 2010. Shenwei is a military project. It not only does not have to face the most troublesome problem of technical brain drain of IC design companies, but also can be fully guaranteed by the military in terms of human, material and financial resources.
Secondly, Shenwei focuses on the supercomputing field, and there is no dilemma of lack of market competitiveness caused by poor software ecology and weak industrial alliance encountered by Godson in the PC market. Shenwei has its own instruction set, microstructure, CPU, compiler and operating system, which is enough to gain a foothold in the field of supercomputing. Godson is constrained by the lack of software ecology and weak industrial alliance. However, Shenwei's focus on supercomputing also brings a regret, that is, in the civil PC market most closely related to the people, we may never see Shenwei's products.
Finally, Shenwei is a military project, and Shenwei supercomputing is a national project. Shenwei's software and hardware products can completely complete the internal cycle within the system. Many of Godson's products must fight in the commercial market.
Since its birth in 2003, Shenwei has independently expanded its instruction set based on alpha, designed five chips of Shenwei 1, Shenwei 2, Shenwei 1600, Shenwei 1610 and Shenwei 5, independently developed Shenwei smart compiler and Shenwei Ruisi operating system, and launched PCs, servers, high-performance computers, supercomputers, firewalls, rack storage servers Large scale cluster storage system, thin client, QDR HCA card, Shenwei cloud and other products have done their best for national information security and supercomputing construction, and successfully completed the tasks delivered by the state.
The official account of WeChat is searching for "love plate network". With the latest development boards, intelligent hardware, open source hardware and activities, you can master all your hands. Recommend attention!
[wechat scanning can be followed directly]
Related reading:
Xilinx starts shipping 16nm FinFET chips
Melexis programmable multi-channel RF transceiver aims at low-power sub GHz applications
China's first aid
Our other product:
