Supercomputer is a super-large electronic computer. Supercomputer is an important tool to obtain ultra-high computing performance and solve large-scale scientific computing and massive information processing problems. Supercomputers are used to perform high-precision and high-speed numerical calculation simulations to solve the problem of solving the problem of high-precision and high-speed numerical calculations. The role played by practical problems is unparalleled. The challenging problems that rely on supercomputers to solve cover every field of human science and life, including nuclear test simulation, petroleum exploration, meteorological forecasting, agricultural breeding, medical services, new drug development, animation rendering, material design, financial computing, etc.
The United States banned the sale of supercomputers
964, the world's first vector supercomputer CDC6600 was unveiled. This behemoth with 350,000 transistors installed has a computing speed of 3 million times per second, 10 times that of other computers.
975 began to develop supercomputers. After 8 years of long development, " Galaxy 1 " was finally born. China became the third country in the world to develop supercomputers, making the supercomputer family "Chinese faces" appear for the first time. : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : "" ".
In the following decades, this "face" was constantly refreshing, and he contributed his wisdom to the world supercomputer family again and again.
009, " Tianhe No. 1 " was born. Tianhe No. 1 broke through a series of key technologies and creatively adopted a multi-array, configurable, collaborative and parallel architecture. Its peak operation speed reached 4700 trillion times/second. This is my country's first supercomputer with a 10 trillion-times time, and it is also the second country after the United States to develop a 10 trillion-times time. It can be said that it has broken the long-term monopoly of the United States.
In 2010, China's "Tianhe No. 1" Phase II system ( Tianhe -1A) ranked first in the latest global supercomputer top 500 rankings. It calculates 1 hour, which is equivalent to 1.3 billion people calculating for more than 340 years at the same time.
China's development on supercomputers has caused concern in the United States. In April 2015, the US government announced that it would ban the sale of "Xiangma" (XEON) chips to four national supercomputer centers in China, with the reason being that two Intel micro-type models were used. The processor chip's Tianhe 2 system and the earlier Tianhe 1 A system, "believe it is used for nuclear explosion simulation". In addition, the National Supercomputing Changsha Center, the National Supercomputing Guangzhou Center, the National Supercomputing Tianjin Center and the National University of Defense Technology were included by the United States on the "list of entities that insist on violating US national security or diplomatic interests."
China breaks through the blockade and blooms in full swing
The US ban has banned China's possibility of continuing to obtain Intel high-performance chips. But it also promoted the further development of Chinese supercomputers. In fact, long before the US sanctions, Wuxi Jiangnan Computer Research Institute had long anticipated that the United States might restrict the development of Chinese supercomputers and began to develop its own chip - Shenwei .
Jiangnan Computing has selected the decline of Alpha architecture, and the instruction set is also extended based on Alpha. The early technology source of Shenwei was from DEC's Alpha 21164. DEC has strong technical strength, but its business model is not as good as others and failed. Alpha has been replaced several times, and later it was even more shelved. The instruction set and microstructure are no longer updated, and most technical patents have expired or are about to expire. This is also the reason why Jiangnan Computing chose the Alpha architecture - the United States has no way to support it. Shenwei is the only remaining fruit in the Alpha camp at present, and has the autonomy to independently expand instructions and development routes.
Shenwei uses the Alpha instruction set as the basis, expanding its own instruction set, and designed Shenwei 1 with independent microstructure in 2006. Shenwei 1 is a single-core CPU with a 130nm process technology with a main frequency of 900MHz. Integrated 57 million transistors. The Shenwei 2, completed in 2008, is a dual-core CPU with a 130nm process technology, with a main frequency of 1.4GHz.
and Shenwei 1600 is the third generation processor developed by the Shenwei series developed by the Jiangnan Computing Institute. It has built-in 16 RISC 64-bit cores, 65nm process technology and 975-1200MHz operating frequency, and has a floating-point computing capability of up to 140.88GFLOPS (1.1GHz). Working frequency performance), equipped with Shenwei Blu-ray ultracomputer, and also uses the domestic Shenwei Ruisi operating system, realizing the domestic production of software and hardware.
Shenwei Blu-ray Ultracomputer has a total power consumption of only 1074KW, which is one quarter of Tianhe's power consumption of 4 MW. Shenwei Blu-ray LINPACK was 74%, which was comparable to the Jaguar Supercomputer, the fastest computer in the United States at that time.
Later, Chengdu Shenwei Technology launched the fourth generation Shenwei chip - Shenwei 26010. Shenwei 26010 does not have an Alpha instruction set. The earwax uses its own developed instruction set, which is specially developed for high-performance computing Instruction set.
Shenwei 2610 processor has a frequency of 1.45GHz, and the chip structure and performance indicators of the Shenwei 26010 heterogeneous multi-core processor are world-leading. It integrates 4 operation control cores and 256 operation cores. The operation control core and operation core both adopt the Shenwei instruction system. The basic instruction set is compatible. The operation core and operation control core have expanded the 256-bit vector instruction set according to requirements. The operation core and operation control core have expanded the 256-bit vector instruction set
The one equipped with Shenwei 26010 is Shenwei Taihu Light. "Shenwei Taihu Light" is a super-large-scale parallel processing computer system with a computing speed of more than one billion floating-point operations per second, using high-density elastic supernodes and High-performance architecture of high-traffic composite networks. The software system of "Shenwei • Taihu Light" consists of domestic multi-core CPU basic software, parallel operating system environment, high-performance storage management system, parallel language and compilation environment, parallel development environment and other parts. The national production of software and hardware has been achieved.
On June 20, 2016, Shenwei Taihu Light, equipped with Shenwei SW26010, undoubtedly won the world's number one supercomputer position, and achieved four consecutive championships. Shenwei Taihu Light is the first in the world A supercomputer with a peak calculation speed of more than one billion times has a peak calculation speed of 1.254 billion times per second.
In addition to the "Shenwei" series, Chinese supercomputers also need the "Tianhe" series and the "Dawn" series. Tianhe 2 originally used the Intel Xeon E5-2692 12-core processor and Xeon Phi 31S1P acceleration card, due to the US ban, 2017 In September 2000, the Guangzhou Supercomputer Center used the domestic accelerator Matrix 2000 to replace the original Intel Xeon Phi accelerator. Each acceleration card uses 4 Matrix 2000 chips. Each Matrix 2000 consists of 128 cores, with a frequency of 1.2GHz, and can perform 16 double-precision operations per cycle. The peak performance of the processor is 2.45TFLOPS.
And the "Dawn" series supercomputers use the Chinese domestic Dhyana (Meditation) x86 processor manufactured by Haiguang to start production. However, this chip is developed based on the AMD Zen microarchitecture.
It can be said that under the sanctions of the United States, China not only did not regress, but instead achieved flowering everywhere in the development of supercomputer chips, and Chinese supercomputers have achieved national production.
E-level supercomputer-Next finals
At present, China and the United States are fully competing for the next track of supercomputer-E-level supercomputer. E-level supercomputer refers to a supercomputer that can perform 10 billion mathematical operations per second, and is recognized by the world as "the next crown of the supercomputer world." The development of
E-level supercomputer is a cutting-edge field of innovation and competition in the international high-end information technology. The United States, Japan and EU have all proposed development plans for E-level supercomputers, almost all of which will be completed around 2021. (Due to the epidemic, it is estimated that it will be until 2022)
At present, China has developed 3 E-class supercomputer prototypes, namely Shenwei E-class, Shuguang E-class and Tianhe No. 3 , Shenwei E-class is completely independent The Shenwei Ruisi operating system, Shenwei Ruizhi compiler and other developed.All computing systems use "Shenwei 26010+" multi-core processors, and all high-speed interconnection network systems use Shenwei network switching chips and Shenwei message processing chips. These key components have completely independent intellectual property rights. The storage and management system is built by Shenwei multi-core processor to achieve a national production replacement of products in this field.
while Tianhe No. 3 independently designed three chips: "Maichuang" multi-core processor (Matrix-2000+), interconnection chip, and router chip, realizing the independent control of the computing system. In 2021, the "Tianhe" E-level (10 billion times) computer key technology verification system won the world's first place on the SSSP Graph500 (single source shortest path) list and the world's first place on the BIG Data Green Graph500 (big data graph computing energy efficiency) list A good achievement.
At present, the United States is also making efforts on E-level supercomputer . In order to achieve its leading advantage over China, the US Department of Energy has previously announced that it will build two E-level supercomputers "Aurora" and "Frontier" with floating-point operations The speed will exceed 10 billion and 15 billion times per second, respectively.
Let us also look forward to China's ability to surpass the United States again in this supercomputing war between the United States and Japan!
