Supercomputer, supercomputer, is the fastest and most powerful class of computers. It plays a pivotal role in cutting-edge fields such as scientific research and military. It is a manifestation of comprehensive national strength and is a national weapon.
These high-end behemoths are far away from the lives of ordinary people, so we can't even perceive their existence. But recent events and some news have stung many people's nerves: The United States recently released the 'Summit' supercomputer, after five years, finally regained the 'world's first supercomputer' throne from China's 'Shenwei·Taihu Light'.
What? We have been leading the United States for five years in such a high-end. Some small partners may have such a sigh when they see this news. Yes, the answer is yes. Out of curiosity, the author follows This news thread has found a lot of interesting data and interesting stories in this strange field. I dare not decide to share it and share it with everyone.
Through the role of Summit, let's first carry out some small science knowledge. How do you calculate the amazing computing power?
The Summit supercomputer is led by IBM. The entire super system consists of 4,608 compute nodes, each of which contains two 22-core IBM Power9 processors (for coordination) and six NVIDIA Tesla V100 graphics processing unit accelerators (for computing). provide Up to 20 billion floating point operating speed spikes per second!
So, how do you understand this 20 billion floating point operations per second? If you use our common laptop as a reference, The Summit supercomputer has the equivalent computing power of 1 million notebooks; if a person can perform 1 calculation per second, he needs to live more than 6.3 billion years to be comparable to the Summit's every second.
The amazing power of Summit is naturally a big guy. Each of the 4608 nodes is like a refrigerator. The total weight is 340 tons. It is placed in a large room of nearly 900 square meters. It is said that the cables connecting these nodes are close. 300 km in length.
Because the power is amazing and natural, the heat is naturally huge, so all use water-cooled heat. Summit runs 1 million liters of coolant per minute, and it consumes enough electricity to provide lighting for more than 8,000 American homes.
Since we know that the strength of the US Summit Supercomputing is good, then what is the level of China's super-calculation?
02 Global super-calculations, China's strength is not weak
For the ranking of global supercomputers, there is a well-recognized 'TOP500' list in the world, which is jointly compiled by foreign authoritative super-calculation experts and has high credibility in the world.
In the 2018 'Super TOP500 List', beauty The latest release of Summit and Sierra has been ranked 1st and 3rd in the list, while China’s 'Shenwei·Taihu Light' and 'Tianhe 2' have fallen to the 2nd and the 4 positions.
And if you look closely at this list, you will find that their rankings are sorted by Rmax (measured peak) and Rpeak (theoretical peak). For any computer, Rpeak theoretical performance is more References, Rmax measured performance is even more critical, even for supercomputers. Next we narrow the scope to TOP5, to see how big the world's top supercomputers are in terms of strength.
The world's top super TOP5 strength ranking
The top spot on the previous list was held by China's Shenwei·Taihu Light for several years. The United States released Summit after five years and regained its first throne.
From the data point of view, Summit's measured peak value is about 31% higher than that of Shenwei·Taihu Lake. Since we use the measured peak data to compare and consider the difference between the two creation times, the actual difference between the two is actually not Like some media claims, the gap is almost double.
China’s super-power is still the strongest in the world
The next set of data will be more interesting. Previously, the media had classified all the supercomputers in the TOP500 list, and found that the number of supercomputers from China was 206, with a ratio of 41%.
The number of supercomputers from the United States is 124, with a ratio of 25%. The two superpowers in China and the United States dominate more than 65% of the world's supercomputers. Although individuals are surpassed by the United States, the scale of China's supercomputer is still the first in the world. .
So the question is, how can China once again surpass the US Summit to regain the top spot in the world of super-performance? Some small partners say that more processors are better, but the real situation is not so simple!
03 China's 'core' forces emerging under the technical blockade
Building a complete set of the world's top supercomputing systems is not as simple as stacking processors. First, you must have powerful enough computing chips, advanced computing management technology, operating system support, and sufficient financial support to make it possible. Of course, behind all of this, the core factor is that there must be policy support at the national level, as well as sufficient cutting-edge scientific and technological talent reserves. Therefore, super-calculation is a manifestation of a country's comprehensive strength.
The super-country of the country started relatively late, but it is progressing rapidly. This is inextricably linked with China's rapid development needs and national policy support. China's Tianhe No. 2 supercomputer has won the world's top TOP500 champion six times and established China in the world. Absolute status in super-calculation.
However, due to the intel Xeon chip solution used in the early days of Tianhe II, it attracted some questions and criticisms. The rapid rise of China's supercomputer also caused the US government to issue a sensational 'restricted sales' event in 2015: The US government prohibits companies such as Intel from exporting high-performance computing chips to four Chinese supercomputer centers..
Domestic Matrix-2000 accelerator replaces intel Xeon
In 2015, the US 'restricted sale' incident broke out, and China decided to replace it with domestic chips. Until September 2017, Guangzhou Supercomputer Center announced that it would upgrade the Tianhe No. 2 supercomputer system before the end of the year. The original Intel Xeon Phi accelerator was replaced by the domestic accelerator Matrix-2000. The Tianhe 2A supercomputer upgraded by the domestic chip achieved an amazing double of the computing power.
Domestic Matrix-2000 doubles the performance of Tianhe 2A
The emergence of the domestic Matrix-2000 broke the 'technical blockade' in the United States. The floating point performance of the upgraded Tianhe 2A supercomputer has increased from the previous 54.9PFLOPS to 94.97PFLOPS, and the performance indicators of interconnection and I/O have also gained a large margin. Technological progress.
The network bandwidth has been upgraded from the original 10Gbps to 14Gbps, the delay has dropped from 1.57us to 1us, the memory capacity has been upgraded from 1.4PB to 3.4PB, the storage capacity has been upgraded from 12.4PB to 19PB, the bandwidth has doubled to 1TB/s, and the power consumption has been reduced. 17.8MW fell to 16.9MW, and energy efficiency increased significantly.
The slightly mysterious Matrix-2000 chip is independently developed by the National Defense Science and Technology University, and the network transmission is produced by Tianjin Maichuang Company. In China, the super-initial self-inflicted internal worries, superimposed on the external situation of the US 'banned sale', the Chinese super-calculation With the self-developed Matrix-2000 to achieve 'Jesus anti-kill', this move can undoubtedly go down in history.
04 Only a firm and completely independent research and development is the way out
Since the implementation of the China 863 Program, the research and development of supercomputing systems has been highly valued by the state. However, due to the weak foundation, it has been subject to people in the international arena.
In 2015, the US Department of Commerce implemented the 'banned sale' high-end chip event, which strengthened the route of independent research and development of high-end chips in China and was promoted to the national strategic level. China's most powerful supercomputer now - Shenwei·Taihu Light It was born in this context.
The Shenwei·Taihu Light Supercomputer installed 40,960 China-developed 'Shenwei 26010' many-core processors. The multi-core processor uses a 64-bit independent Shenwei command system with a peak performance of 1.25 billion times per second. Performance is 9.3 billion times / second.
In 2016, the list released by the international TOP500 organization showed that the 'Shenwei·Taihu Light' supercomputer system topped the list, not only twice as fast as the second 'Tianhe II', but also improved its efficiency. 3 times.
The Shenwei 26010 many-core processor uses a 64-bit independent Shenwei command system with a peak performance of 3.168 teraflops per second and a core operating frequency of 1.5 GHz. There are 40,960 such processors installed in the Shenwa Lake Light. Using the 'Shenwei Taihu Light' made by China's own chip to replace 'Tianhe No. 2' to the top of the list, China's total number of super-calculations has surpassed the US for the first time in history.
The emergence of the Shenwei 26010 many-core processor basically achieved the replacement of the Intel Xeon Phi accelerator card. Although there is still a performance gap between the performance of the single-chip and the Intel Xeon, but thanks to the advantages of size and power consumption, It can make 256 computing nodes in a small chassis, and the same volume computing performance can kill Intel Xeon Phi.
Compared with other domestic chips, Shenwei has to be more low-key and mysterious. Because Shenwei processor has been carrying the mission of military defense and national information security since its birth, the 56th Research Institute of the General Staff of the R&D unit. (Wuxi Jiangnan Institute of Computing Technology) In 2003, it began to design Chinese own high-performance chips. Under the support of the national 'nuclear high base' major special project, Shenwei processing was developed using autonomous instruction set, and has completely independent intellectual property rights.
Although China's two heavyweight supercomputers 'Shenwei·Taihu Light' and 'Tianhe 2A' are surpassed by 'Summit' and 'Sierra' from the United States, it is reasonable to have two strong rivals.
According to the latest news, China's 'Tianhe No.3' is actively researching and developing the public, and is expected to regain its first throne in 2020, and achieve a performance that exceeds eight times the performance of 'Shenwei·Taihu Light'. The 'Tianhe No.3' is expected to adopt a fully independent innovation design, autonomous Feiteng CPU, independent Tianhe high-speed interconnect communication, and an independent Kirin operating system.
However, we must be soberly aware that the opponents we face in the chip field are so strong, regardless of IBM, Intel, nVIDIA, any chip company is strong enough.
At the same time, it is important to note that as early as five years ago, the US government stopped funding the sponsorship of supercomputer projects. Instead, it began to invest heavily in the research and development of quantum computers. This was also the case that the US supercomputer stopped before and was eventually China. An important reason for surpassing one fell swoop.
The challenges and missions placed in front of domestic chips are equally important. Under the current international situation, only the path of independent research and development, which is identified, is completely the only way out of being controlled by talents.
Whether it is the 'Tianhe No.3' that will be put into use in 2020, or the follow-up upgrade of 'Shenwei·Taihu Light', in the field of supercomputers related to the national economic lifeline and information security, domestic chips will grow stronger, let us wait and see !
A feature film on Shenwei 26010
