Ring time: The US embargo has caused China's next supercomputing to be fatal?

In recent days, many American media have gloated and believed that China's supercomputing industry will be killed by fatal blows. Without the CPU in the United States, China will not be able to create supercomputing.Is this the case?What impact does this ban have on my country's supercomputing industry?

According to multiple media reports, on June 21, local time, the US Department of Commerce once again included 5 Chinese entities in the entity list of entities that are regulated by import and export, and they are prohibited from purchasing components from the United States suppliers.Related decision to take effect on June 24.The five entities are related to the two of the three giants in China's supercomputing industry, involving the two supercomputing of Shuguang and Shenwei.The Jiangnan Computing Technology Research Institute in these five entities is the research and development and manufacturing unit of Shenwei Supercomputing.The other four entities, the Dawn of Zhongke, Tianjin HIGON, Chengdu Haiguang Integrated Circuit, and Chengdu Haiguang Microelectronics technology are all related to the dawn series.

First of all, why did the United States sanction in my country's supercomputing companies again?

In fact, as early as 2015, the US Department of Commerce included China's four super -computing centers on the embargo list.This stems from the long -term anxiety in the United States: In about 40 years, China's supercomputing industry has been fully imported from American products to purchasing the US CPU to make supercomputing by itself, and then using its own CPU and accelerator to manufacture super computors; from onlySeveral supercomputing units are not worth mentioning in the world's supercomputing market, and to the world's number one in the world, which greatly surpass the United States; from low computing properties that have low computing properties, it is not available to the world's fastest supercomputingThe first place in the rankings is many years; China's supercomputing industry has embarked on a perfect counterattack.

The World Supercomputing Organization TOP500.ORG launches the world's fastest 500 supercomputing rankings every six months.The latest list just launched in June this year.The fastest world's fastest supercomputing quantity is as follows as the country's distribution:

▲ The data comes from TOP500.ROG, compiled by the author

In the TOP500 list, the number of overhauls in China has reached 219, accounting for 44%of the world's, and the world's first; the United States accounts for 23%of the world's number of units, which is much backward than China.

You know, in June 2018 just a year ago, China ’s total supercomputing in TOP500 is still 206, and the United States is still 124.In this year, China has made progress again, and the United States is regression.From the perspective of trends, China's advantages will become increasing.

Due to the huge role of super computing in military, scientific research and economy, the anxiety in the United States is understandable.They were anxious to restrain China's development momentum, so they still put the principles of business integrity and fair trade into the sewer and sacrificed the embargo.So, how does this embargo have on China's supercomputing industry?

There is no impact on the three E -Class supercomputing in China in China.

In the summer of 2018, the U.S. supercomputing Summit finally ranked first in TOP500, squeezing the light of China Taihu Lake for many years, and it was about 60%faster than the light of Taihu Lake.A super -compassion Sierra was put into use, and it squeezed the light of Taihu Lake to third place with a weak advantage.

The media in the United States became excited, changed the silence of many years, and began to brag about the power of the next -generation E -class supercomputing.The so -called E -class supercomputers refer to the supercomputers of 10 billion yuan per second, which is much better than Summit.The E -Class supercomputing plan in the United States is planned to be put into use around 2021, and China has prepared three solutions for the E -Class supercomputing. All of the domestic CPUs and accelerators are also planned to be put into use from 2020 to 2021.

Here you need to introduce the accelerator. The core component of the supercomputing is not just the CPU. Because the supercomputing is densely scientific computing work, the CPU performs such computing tasks is not the most efficient.High, the typical accelerator chip is the universal GPU of NVIDIA, and the Xeon PHI processor of Intel.

In recent years, there is a trend that has appeared in the supercourage in recent years. It is to use a CPU with multiple accelerators. The computing task is mainly completed by the accelerator chip. The CPU is mainly coordinated, communication and distribution tasks.

So, how does this embargo in the United States affect our E -Class supercomputing?Let's take a closer look.

The three sets of the next generation of E -Class supercomputing in my country are all using domestic CPUs and accelerator chips:

Tianhe No. 3 E -Class Super Calculation:

On May 19, 2018, the National Supercomputing Tianjin Center demonstrated the prototype of Tianhe No. 3 and was put into use at the end of 2018.The full configuration host of Tianhe III is planned to be put into use in 2020, and the performance plan is 1EFLOPS.Tianhe III will use the domestic Feitan FT2000Plus or the CPU of the follow -up model, as well as the Matrix2000Plus of the University of Defense Technology or the follow -up model accelerator chip.

Feiteng CPU adopts ARM architecture.

The Matrix2000 acceleration chip is a general DSP chip of the National Defense University of Defense. GPDSP is just like Nvidia General GPU (GPGPU). DSP (digital signal processor) also has a large number of computing units.

The supercomputing of the Shenwei series was developed by the Jiangnan Computer Research Institute.In July 2018, the E -Class prototype of Shenwei was settled in the National Supercomputing Jinan Center. It will take about three years of research and development, and will be officially put into use in 2021.Shenwei's CPU is definitely the latest model of the Shenwei series, and the accelerator may also use Matrix2000Plus or its subsequent models.

The Shenwei series CPU was born from the DEC Alpha processor architecture.

Shuguang E -Class Super Calculation:

Zhongke Shuguang mainly uses the X86 architecture CPU.In the latest TOP500 list, there are 63 supercomputers made by Shuguang Company.In addition to one, the other 62 units use the Intel's most powerful CPU, and some supercomputers also use Nvidia TESLA accelerator chips.

A fierce look, it seems that the embargo is a fatal blow to Shuguang.Because Intel and Nvidia chips will not be available in the future.However, Shuguang has long had the national production layout.The secret is on the super compassion outside these 62 units.

CPU:

This model of TC8600 was used in 2018. Its CPU uses Tianjin Haiguang Company's Hygon DHYANA processor, 32 cores, and 2GHz frequency.Tianjin Haiguang Company obtained the permanent authorization of the first -generation Zen architecture of AMD in the United States in early 2016. The starting point is quite high. You must know that AMD launched the second -generation Zen2 architecture in 2018.Haiguang 3000 series CPU uses the X86 universal architecture, which fully supports mainstream operating systems such as Windows 10 Pro, Windows Server2019, Linux CentOS 7.X, andThere is also a server version similar to AMD EPYC, with a maximum of 32 cores and 64 threads, a basic frequency of 2.0GHz, a acceleration frequency of 3.0GHz, and covering multiple product lines from 4 to 32.

After this embargo was announced, Haiguang could no longer be authorized by AMD's new product architecture, but this did not affect the authorized authorization.Haiguang has formed a R & D team of more than 500 people. It has the ability to further develop new products and improve the performance of the CPU on the basis of the authorized authorization.This also means that the CPU of the X86 architecture may produce a hard fork in China in the future. We will separate from the United States and come out of our own way.

Accelerator chip:

And this dawn supercomputers, the accelerator chip is more interesting: its name is Deep Computing Processor.According to public news reports, Shuguang's accelerator chip was developed based on the AI processor architecture of Cambrian Corporation in my country, which is unique in the world.Shuguang's draft said: In November 2017, Shuguang was the first to launch Phaneron based on Cambrian low power deep learning processors, which attracted widespread attention in the industry.After 6 months, the Cambrian released MLU100, which has a stronger performance. Shuguang upgraded Phaneron products simultaneously, and released a series of server products equipped with Cambrian MLU100, maintaining a continuous leader in the field of smart server

This means that Shuguang TC8600 supercomputing will have great advantages in the AI field.In the future Shuguang's E -class supercomputing is the latest version of the latest CPU and Shuguang's latest CPU and Shuguang.

It can also be seen that the embargo from the US Department of Commerce cannot affect Shuguang's E -Class supercomputing product at all.

The embargo may have a huge impact on Shuguang's commercial supercomputing product line

We also need to see that because Shuguang's main product lines are based on Intel's CPU and Nvidia's accelerator chip. After this embargo, these product lines are difficult to continue.From a business level, the impact on Shuguang Company is huge.Fortunately, Shuguang has a national CPU and accelerator chip replacement product.We can see some details from the two supercomputing data of the 43rd Shuguang TC8600 and the 88th Dawn TC4600LP/W740L.

Shuguang TC4600LP supercoute is a product based on the Intel -Qiangqiang CPU and Nvidia Tesla accelerator.Judging from the above test data, the TC8600 super calculated using the domestic CPU and accelerator, in the case of more than 3 times the number of CPU cores and about 5 times the core quantity of the accelerator, the running score has only increased to 1.4 times to 1.7 times to 1.7 times.EssenceHowever, the power consumption has decreased a lot, and the power consumption efficiency has been increased by more than three times.

The author analyzes this mainly because the AI accelerator used by TC8600 is not suitable for traditional floating -point operations. It just has a huge advantage in AI operations.

This also means that Shuguang lacks a suitable accelerator chip to improve the floating -point computing power of the system.This may adversely affect the commercial development of Shuguang Company.Maybe Shuguang's future sales focus will be shifted to AI operations.However, it does not rule out that Shuguang will develop products based on domestic Matrix2000 acceleration chips.

Here we must introduce why Shuguang cannot give up the X86 architecture CPU.

100%of the global TOP500 supercomputers use different distribution versions of the Linux operating system.These versions are optimized for their respective CPUs and accelerator chips.The application software running on the supercomputing is also affected by the CPU architecture.

From the current point of view, the linux system optimized in the X86 is the most mature, the most software is the most, the easiest program development, and the entire software ecosystem is the most mature and stable.The system of domestic Shenwei and Feiteng CPUs is somewhat insufficient. Software tools optimize less and have a little trouble in programming. There are more system bugs. The entire software ecosystem is only solved.A certain distance.

Therefore, in terms of national strategy and from the business prospects of Shuguang Company, the X86 architecture should be done.

The advancement of Chinese semiconductor manufacturing technology will greatly promote the performance of domestic supercomputing CPUs and accelerator chips.

China's semiconductor manufacturing industry has been our biggest shortcoming for a long time.When TSMC and Samsung Electronics had mass production of 7 nanometer, Intel had a small -scale mass production of 10 nanometers, my country ’s most advanced semiconductor manufacturing process is still SMIC's 28 nanomal process.

The domestic CPU mentioned above either uses domestic more backward craftsmanship. The autonomy can be guaranteed, but the performance is backward; either finding a semiconductor foundry enterprise overseas can improve the performance, but the autonomy cannot be guaranteed.Time will be stabbed around the neck.

But this state will soon end.SMIC announced in the industry report in August 2018 that its 14 -nanometer production process has made significant progress. The first generation of FinFET technology research and development has entered the customer introduction stage.The formal mass production is planned to be in mid -2019.So, soon, we will have our own 14 -nanometer semiconductor production technology.

How powerful is 14 nanometers?You know, all desktop and server chips in Intel are now 14 nanometers.Intel announced at the Taipei computer exhibition just past the small -scale mass -produced low -voltage laptop processor, and it is expected that 14 nanometers will still be its main process in the next few years.Only AMD was under the foundry of TSMC, and just announced that the CPU's 7 -nanometer production was achieved.

If domestic supercomputing CPUs and accelerator chips are produced in SMIC's 14 nanometer, their performance will get a leap.

The United States will eventually move a stone to smash its own feet

In fact, the embargo of the four major supercomputing centers in China in 2015 has smashed its own feet: Intel's strongest PHI accelerator chip was discontinued.

I list some facts here:

China ’s Tianhe No. 2 overhaur used to use 32,000 Xeon E5CPUs and 48,000 Xeon PHI accelerators.For Xeon PHI's sales, Intel CEO said that the goal of 2016 was 100,000 shipments.It can be seen that my country ’s Tianhe No. 2 can reach about half of the Xeonphi sales of Intel in 2016.

On April 9, 2015, the US Department of Commerce announced that the four supercomputing centers in China embarked on Intel's Xeon processor and the dedicated PHI accelerator chip.

On June 18, 2016, the price of the Soluble PHI7290/7250 accelerator was cut directly.By August 24, 2017, Intel announced that the Muso PHI 7200 Series Association processor acceleration card entered the process of discontinuation and delisting.

▲ The picture shows the importance of American personal computer magazines in the report of the year that the importance of the Chinese market for the Dable of Intel, and the US ban will bring a heavy blow to Dette

The author has no doubt that the discontinuation of the Intel -Qeon PHI accelerator chip is due to the embargo on China.This even affects the development of the next -generation E -Class supercomputing in the United States.The E -Class supercomputing in the United States, the CPU can choose the X86 processor of Intel and AMD and the Power processor of IBM, and the accelerator chip currently has only one selectionIt is Nvida's GPU.

So what are the results of this embargo by the US Department of Commerce?The author believes that with our self -reliance, a few years later, several new supercomputing CPU ecosystems will appear in the world to completely break the American monopoly.

In fact, neither the New York Times and the Wall Street Journal of the United States also believe that the US government's campaign to combat China's supercomputing will hinder China's development.The two media interviews several different US supercomputing experts said that although the behavior of the US government has hindered the development of China's supercomputing in the short term, it will also strengthen China's determination to develop local technology to replace US technology.

And one thing that will inevitably happen is that in 2020 and 2021, three E -class super computors in China will impact the championship position of the TOP500 list and kick the US supercomputing.Let the United States look at who is the king of supercomputing!

Ring time: The US embargo has caused Chinas next supercomputing to be fatal?