Daily troops to answer your questions
First of all, the computing power of a supercomputer does not depend on the CPU in the narrow sense. What a supercomputer needs is floating point computing power. At present, Commercially available CPUs have inherent shortcomings. In Tianhe-1 and Tianhe-2, what really provides computing power for the supercomputer is the GPU and co-processor, just like your PC, CPU scheduling, and graphics card work. During Tianhe 1 and 2, the hardware was purchased directly. Tianhe 1 and Tianhe 2 basically used Intel xeon E5 xeon phi, and even AMD graphics cards (the real graphics card hd4870x2, not even the computing card) ). After that, the United States began to restrict exports to China, mainly embargoing products such as xeon phi and nvidia Tesla, but CPUs were not so strict. TaihuLight has begun to use self-developed processors. The design idea of ??sw26010 is essentially similar to the Intel lga interface xeon phi. The many-core stack computing power, as for the process, is 28nm. This process was placed in 2015, it can be said to be a mainstream process, and it is by no means backward. , and the energy efficiency wire is several grades better than Intel. Of course, this kind of processor makes no sense for PC.
What’s more interesting is that the xeon phi 7200 series, which has a similar design idea to sw26010, was only launched this year. So, let’s first ask if it is true, and then ask why. Of course, Chinese CPUs are good, and they are very good.
Then why not for civilian use? Without a patent or authorization, it cannot adapt to the market.
In order to produce CPUs, in addition to paying relevant patent fees to foreign countries, my country only has to establish a set of communication patents and protocols exclusive to itself. The key is that these two things must be recognized by other computer suppliers. Otherwise it's useless.
Peter Beckman, director of the Academy of Science and Engineering at Argonne National Laboratory, once said in an interview with the media, "Over the years, China has The investment and efforts in supercomputing are huge, and the development speed and achievements are amazing." Moreover, colleagues from abroad have also begun to envy China, praising China as "the country is doing things" (which means that China is very efficient in doing things). Chinese supercomputing systems, namely "Sunway TaihuLight" and "Tianhe-2" still occupy the top two positions in the latest TOP500 list. To a certain extent, China’s achievements in the field of supercomputing are highly related to its industrial, academic and military capabilities. The people of the country are proud not only because "Sunway TaihuLight" once again occupied the top spot in the TOP500 list without any pressure, but also because it is about five times higher than the fastest supercomputing system "Titan" in the United States. The speed is also because "Sunway TaihuLight" uses chips with China's independent intellectual property rights. Moreover, "Sunway TaihuLight" has achieved unprecedented breakthroughs in key technologies such as architecture, high-speed interconnection and system software. "Tianhe-2" uses Intel chips.
Sunway Blu-ray computer system is equipped with the domestically produced "Sunway 1600" central processor.
Many domestic industry players believe that Chinese people should not be complacent about China’s achievements in the field of supercomputing. Currently, the U.S. Department of Energy is supporting the construction of two new supercomputing systems, "Summit" and "Sierra." The supercomputer "Summit" located at the Oak Ridge National Laboratory in the United States is expected to be put into use in 2018. The computing performance of "Summit" will be about twice that of the "Sunway TaihuLight". In other words, the United States will regain its dominance in the supercomputing field in 2018. Moreover, American companies such as Intel still dominate the chip industry. For example, in the latest TOP500 list, 471 supercomputing systems use Intel chips, and 14 supercomputing systems use chips from International Business Machines Corporation.
On this topic, some media directly quoted Ji Qing, deputy general manager of the high-performance computer product division of China Sugon Company, as saying, "E-class supercomputers are the crown jewels that countries around the world are competing for, and the occupying strategy The commanding heights are crucial, but from an industry perspective, our more fundamental original intention is to provide more choices and possibilities for global users.” Frankly speaking, Ji Qing said it very well. China's development of supercomputing systems is not to crush its foreign counterparts, but to serve people around the world.
China still has a big gap compared with its counterparts in the United States and other countries in the chip field. What is the reason? I searched on many websites at home and abroad, and finally found some words said by Hu Weiwu of Loongson Zhongke in an interview with the media in early 2018, which specifically included three points.
First, if we use a race as a metaphor to compare the competition between countries in the field of chips, in the past China could not even see the backs of its foreign counterparts. Today, at least it can clearly see the backs of its foreign counterparts. Well, wait until 2020 and you will be able to see the back of the head and hair of your foreign counterparts clearly. China is chasing hard, but its foreign counterparts are also running hard ahead. For China to catch up with its foreign counterparts, there must be a process. Second, it has been almost 40 years since China's reform and opening up. In these four decades, China has made many remarkable achievements. However, in the past, China directly bypassed the development of high-precision technologies such as general-purpose chip CPUs, operating systems, engines and high-end instruments. Now, if China wants to grow from big to strong, it must go back and conquer these high-precision technologies. It will be a long process for China to capture these sophisticated technologies. Because these high-precision technologies are very complex, there are many factors that affect these high-tech products. If China wants to develop these complex systems, it can only proceed step by step. Third, in the process of developing complex systems such as general-purpose chips, operating systems, engines, and high-end instruments, China is not doing so because of a lack of money, nor is it because of major institutional problems. The most important thing is a lack of time. Moreover, if China wants to develop chips, operating systems, engines and other products that can be mass-produced on a large scale and have international competitiveness, it must constantly discover problems during the application process and then continuously improve products in the process of solving problems. .
To put it in the simplest sentence, China’s foundation in high-tech fields such as chips is much weaker than that of its foreign counterparts, which has led to China lagging behind its foreign counterparts in the field of chips. In fact, today, in many high-tech fields, China is playing the role of a catch-up. Let me tell you, if China had not missed the first, second and third technological revolutions, China's achievements in the high-tech field today would definitely be much greater and more brilliant than it is now. We should not investigate the history of more than 100 years before the founding of New China. What's more, based on the speed of development China has achieved in the past 40 years of reform and opening up, China will definitely become one of the world's major high-tech innovation centers in another 20 to 30 years and benefit all mankind.
Domestic CPUs are not lagging behind. What lags behind is the CPU ecology under the domestic intellectual property system.
The brilliance of China’s supercomputer, especially the Tianhe system, comes from the heterogeneous computing proposed by the Chinese designer team for the first time in the world, that is, the joint work of CPU and GPU. Of course, the Tianhe series mainly used Intel chips before. Later, after the U.S. Department of Commerce restricted exports, the domestic Sunway chips immediately took over and continued to maintain the global dominance of Chinese supercomputers in the Sunway TaihuLight supercomputer. Leading position.
The backwardness you see now comes from poor user experience. This difference does not come from scattered hardware, but from the immaturity of the entire system.
It’s like if we take the engine of Rolls-Royce, the aerodynamic shape of Lamborghini, the transmission of Transchi Blue Sky, and the chassis of Mercedes-Benz G, and build a car, will this car crush the car? Overwhelming the world?
Not really. Because these systems are not compatible. Such a Chimera sedan may not even be able to outrun the new Civic that is clamoring on the streets.
Domestic CPUs were originally a lonely and healthy heart. But this heart cannot be placed in the body of any living person. To use it, you can only create a person yourself.
Well, it’s not too difficult for us to build one. As a result, this person fell down before he could stand up because his leg muscles did not respond.
Everyone said: Domestic products have such a bad heart.
It’s okay to continue adjusting. Come on.
After the adjustment, the person stood up! Everyone was excited, but after taking two steps, the dynamic stability was not good, and he fell down again.
Everyone said: Domestic products have such a bad heart.
It doesn’t matter, let’s get on with it!
Adjusted well this time! The little man we made is walking! Start running! So exciting! At this time, the Intel platform host that was training to run the 100-meter race passed by like a wind and looked at our villain like a fool.
Everyone said: Domestic products have such a bad heart.
It doesn't matter...we can still hold on.
After completing all the adjustments, our villains stand together with the win system on the Intel platform and the OSX system on the Apple platform. They must ignore the gap in growth time and at least give users Same level of feeling.
We have not yet accomplished this step.
But before this, we have spent more than 30 years of hard work and have completed it.
As for the subject, domestic CPUs are not garbage. Time will tell.
Domestic CPUs are not lagging behind. There are many types of CPUs. For example, the Kirin 970 in mobile phones is the world’s first-tier CPU. The respondent is probably asking about desktop general-purpose CPUs. This is mainly related to the market, not that China cannot develop and produce such CPUs. To put it bluntly, it will be difficult for latecomers to sell similar desktop computer CPUs.
This question is like asking a cook, if your fried eggplants are so delicious, why is your eggplant cultivation technique so poor?
Those who produce computers do not produce CPUs, just like a cook who only cares about cooking but not growing vegetables.
I remember that at a construction machinery exhibition, a media person asked a well-known construction machinery manufacturer: Since the cranes you produce are so powerful, why do you still use imported plunger pumps? Why not conquer this associated device? The manufacturer's answer at that time was that we were making complete machines, not every component had to be produced by ourselves. Engineering hydraulic pumps should not be developed by our assembly plant, but should be developed by manufacturers specializing in the production of hydraulic components.
Super computing is a giant mining machine, and its computing power is mainly realized by GPU. Tianhe-2 uses Xeon CPU and N-card. The self-developed Sunway CPU used by Taihu is said to have 260 cores.
Looking at Taihu’s data, in fact, domestic CPU technology is not that backward. But there is indeed a gap in general-purpose CPUs, because general-purpose CPUs are basically based on the X86 architecture. If you want to develop them, you must ask others to buy a license. Everyone knows that Huawei's self-developed ARM chip Kirin was made by buying out other people's public versions. Is it possible to do it yourself without buying a license? Yes, but you can create a communication protocol yourself and make an instruction set, and others will have to make supporting packages for you. Otherwise, if the motherboard doesn't recognize the memory or the hard drive, it will still be empty. Today's accessories only accept X86. If your U has to adopt this architecture, you must look at the face of the US emperor.
And we all have to buy the version, buy the patent, and look at the face, will the product be an authentic domestic CPU? The supercomputer is not as mysterious as imagined, it is just a calculation tool.
When someone enters the calculation conditions, the supercomputer will output the calculation results. It's the same as the calculator you use to buy groceries, it's just a difference in scale.
Our commonly used home computers generally have only one CPU, and each CPU generally has only 2 to 8 physical cores. This allows you to play games, watch videos, and browse the web, because these programs only require a small amount of floating point operations. If you want to simulate the flow of air near the aircraft and the stress on the aircraft itself, you need a supercomputer with thousands of CPUs to complete it. You need to divide the air and the body into small pieces and calculate each one separately. The movement and force of small pieces are then integrated to obtain the overall movement and force. Supercomputing can divide a large task into parallel calculations, so that heavy computing tasks can be completed quickly.
If you divide a 1 cubic meter cube in the airplane model into small squares of 1 cubic millimeter, you will get 1 billion small squares. An ordinary home CPU needs to perform 1 billion operations continuously, and the entire calculation process may take a whole day. However, if there are 10 CPUs, 1 billion small squares can be divided into 10 points, and each CPU only needs to calculate 1 100 million squares, and then integrate the results of all CPU calculations, and the calculation can be completed in two hours.
Today, supercomputers have been widely used in meteorology and climate, oil exploration, atmospheric and marine environment, aerospace, space simulation, cryptography research, nuclear explosion simulation, weapons development, materials science, industrial design, earthquake simulation, Animation rendering, deep learning, artificial intelligence, biomedicine, genetic engineering, data mining, process control, financial analysis, human tissue system research, public services and other aspects. Therefore, supercomputers have become a battleground for powerful countries. It can be said that in the future, if you want to break the so-called "proton lock" in science fiction movies, you will need to use supercomputers.
The term "supercomputer" does not have a clear definition and changes with the development of the computer industry. Early Control Data Corporation machines were up to ten times faster than their competitors, but were primitive scalar processors. By the 1970s, most supercomputers were vector processors. In the early 1980s, supercomputers developed toward massively parallel computing systems, which consisted of tens of thousands of general processing units. In the mid-1980s, it became a common method to combine 8 to 16 vector processors for parallel computing. From the 1990s to the early 21st century, supercomputer interconnections were mainly based on reduced instruction set tensor processors (such as PowerPC, PA-RISC or DEC Alpha) for parallel computing. The current situation of supercomputers in my country
In the recently released list of the Top 500 global supercomputers, 214 of our country’s supercomputers were on the list, ranking first in terms of number. Sunway TaihuLight and Tianhe have topped the list many times. 2A is located at 4th and 6th respectively. Based on the failure of the supercomputer "Kyo", Japan launched the follow-up "Fugaku" and reached the top again after 9 years, while the American supercomputer "Summit" and "Sierra" were squeezed to the second or third place.
Although we have made great achievements, the CPU is still our shortcoming. For example, in the past, Tianhe-2 used Intel Xeon Phi (Xeon Phi) CPU and topped the Top500 list in the supercomputing industry four times in a row. However, since the United States banned Intel from selling Xeon Phi CPUs to the Guangzhou Supercomputing Center in April 2015, it has obviously been unable to keep up.
Although it can be replaced by 2 to 3 domestic Shenwei chips, using a Tesla V100 to solve the problem will only increase the complexity. The more chips, the more difficult it is to schedule. This is just like a team of 5 people is easy to manage, but the difficulty of managing a team of 500 people increases exponentially. Summary
If my country’s supercomputing road wants to make great progress, it still needs to make up for the shortcomings of the semiconductor industry chain. If Sunway's CPU can also use the 7nm process and HBM2 memory, it may not be shorter than Japan's "Fugaku", and who will win is still unknown. But it is precisely this premise that we are lacking.
What you are talking about is the desktop CPU that is lagging behind. What affects the desktop CPU is the software ecosystem (basically the instruction set ecosystem). To give a simple example, no matter how powerful the domestic CPU is, if it cannot run Windows, will you buy it? ? The reason is that simple, because the instruction set patents for good software ecology are tightly controlled by the United States, so there is no other market, so naturally they will not do it.
Our CPUs are already world-leading, such as Huawei's HiSilicon. But general-purpose processors, such as the X86 series CPUs commonly used in computers, are not something we can do if we want to because of patent reasons. Therefore, the very good processors we make are not commonly used by us in daily life. That is, we can only have access to them if we buy a Huawei mobile phone, and we may not even know about it even if we buy it.
Looking at the past few years, our country has been developing in the past ten years. Which technology has not been accumulated with huge investment?
High-speed rail, large aircraft, manned spaceflight, aircraft carriers... This also benefits from our superior social system, which means spending money to do big things.
In the West, spending big money requires getting approval from states and taxpayers, which is inefficient. For example, if the United States builds a high-speed rail, taxpayers will object if they think there will be no benefits in a short period of time.
Then supercomputers are also built with money. As long as there is a lot of money, it is only a matter of time before they are built. The key is that those big countries are not willing to spend money, which shows how powerful our supercomputers are!
The civilian CPU is not simply a matter of piling up money. Its cost is much lower and the return is much higher. After all, it is mass production. We can also build CPUs. The key problem is that the cost of existing technology is too high. If the cost is high, the price will be high. If the price is high, few people will buy it. If no one buys it, an ecosystem will not be formed. The so-called ecosystem requires thousands of developers around the world to develop on this CPU and promote it together. This is just like Microsoft's mobile operating system. The technology itself is very good, but the ecology is poor. There are no programmers to make mobile APPs on it. There are few APPs and even WeChat, so there are few people buying them... Naturally, it becomes a vicious cycle and it cannot grow. .
The supercomputer laboratory does not rely on the ecosystem. Anyway, it can run on its own, regardless of cost. It only targets high-end users and does not need to consider user experience. It can even use policies to stipulate that certain enterprises and institutions must use it. . This is the biggest difference from ordinary civilian CPUs.