In April 2015, the U.S. department of commerce decided to four national supercomputer center in China to ban xeon PHI calculation card, various bad-mouthing when China super voice be heard. Even the tianhe 2 international supercomputing conference in 2015 topped 5 even the champions league, but by using the Intel CPU and plagued by some people with ulterior motives, many people said that without the CPU, tianhe 2 to die.

At ISC2015 conference, national defense science and technology university announced the tianhe 2 a upgrade scheme, will use the matrix of independent research and development of national defense science and technology university, 2000 (GPDSP) alternative xeon PHI calculation card. Matrix 2000 using 40 nm process, has 16 nuclear, frequency of 1 g, double precision floating point 2.4 TFlops, power consumption is 200 w.

At the beginning of the New Year, from the ministry of public articles, revealed the good news - designed by high performance integrated circuit design center in Shanghai all domestic nuclear chip has reached the advanced world level, the U.S. intelligence community very interested in the chip, so the author calls it all domestic nuclear chips). All the chips with 28 nm process, peak double-precision floating-point operations at more than 3 trillion times a second (3 tflops), completely tied for the second generation of Intel XeonPhi (as well as the core chip of Intel) -- the second generation of xeon PHI calculation code "KnightsLanding" card products, 14 nm technology adopted, double-precision floating-point performance more than 3 tflops, 250-300 w power consumption.

On January 22, the xinhua news agency reported a more encouraging news - China plans to start this year in tianjin new generation ten billion times (ten billion, 1000 p, tianhe of 55 p # 2, if it is not a slip of the pen, is really a black technology) the development of super computer programming... At the same time, the national defense science and technology university is designing a new generation of tang chip.

Thus, the U.S. department of commerce lock-up calculation card behavior didn't have much on the development of Chinese super calculate. In fact, in the past half a century, the history has shown that comrade MAO zedong's words - "block, block ten years eight years, China's problem is solved".

Where can buy it is difficult to do it

Even since the 80 s, China's computer without Chinese core mostly, even the tianhe, tianhe 1 2 these once make Chinese people proud of super computer also use foreign chip, basic domestic chip is used only for high-speed Internet network.

But in the era of MAO zedong, China has its own semiconductor industry, mastered from single crystal preparation, equipment manufacturing, the whole process of IC manufacturing technology, the computer made in China also have a core in China.

But in the 80 s pursued "made than to buy, buy than rent" and "exchanging market with technology", China's integrated circuit industry was devastating, MAO zedong was lost his shirt, accumulated financial technology talents abroad, either erosion or teach at the university, some even to be transferred to guard room, MAO zedong cultivated technology of brain drain.

During this time, autonomous technology is infinite, but YangJiShu by excessive beautification, "foreign experts" is regarded as god, policy-makers alert on its lack even the most basic, a lot of very promising research projects in the lead after listen to the opinions of the "foreign experts" were rejected. A lot of scientific research project is in the "foreign experts" under the "guidance" has been on the circuit; To screw "foreign experts" for technical data, a large number of valuable technology is "foreign experts" stealing has become "YangJiShu", that Chinese companies use their research and development of technology to give foreign investors still pay royalties.

And self-developed CPU also be replaced by imported CPU, such as in 1983, the tide began to adopt imported Intel8088 chip, rather than domestic chip assembly 0520 microcomputer, developed in the early 80 s to the 90 s of the Great Wall, the Great Wall, the Great Wall 386 286 286 2780, 2780, tai chi, the galactic super minicomputers, HN2730 super small computer, such as adopt foreign chip. China's fourth generation, its mainframe computers in the computer represents "Milky Way one" supercomputer at a cost of 100 million yuan, but by buying large quantities of foreign hardware failure to the semiconductor industry in China have much positive progress.

Created in the "buy, buy less than rent" theory, under the guidance of 80 s, China's own technology has been difficult to produce a computer. To the early 90 s, represented by lenovo computer Chinese companies to "MaoGongJi" route, only in low value-added computer assembly, unwilling and unable to be engaged in research and development of the chip, chip market completely in foreign control.

Everything can't buy

After the sino-soviet debate, the United States and the Soviet union technology blockade to China at the same time, to force China to on the development road of independence and self-reliance. As China experts after Khrushchev removed the Soviet union, pushed China with its own ability to get out of the "rocket". In the United States and the Soviet union under the common technology blockade, ha jun gong in 1962 successfully developed the transistor, make China 8 years later than the United States into the transistor era, China's first full transistor computer 441 b - I was born in 1964, compared to the United States in the first full transistor computer RCA501 6 years late. In 1965, China developed the first piece of integrated circuit, than the late 5 years into the era of integrated circuits. In 1972, China developed the large scale integrated circuit four years later than the United States realized from small to large scale integrated circuit IC development.

In China after the loss of the ability of independent research and development manufacturing computer, the U.S. government's strict restrictions on Chinese exports of high performance computer, in addition to the high price of purchase, to put the computer in transparent glass room, monitoring by the americans, the key in the hands of the americans, each use for instructions to the americans, and explain the specific purposes, subject to approval by the americans.

After the painful experience, China's resumption of high-performance computer research and development projects, the galaxy 2 was born in 1992. Dawn dawn no. 1 was born in 1993, no. 1, three days after successfully developed western a high-performance computer to China export restrictions. After established the independent development of super calculate determination, independent research and development of super calculate good news - China has dawn dawn, dawn, 3000, 1000, shuguang 2000 4000, shuguang 5000, shuguang 6000 4, 3, the Milky Way galaxy, the Milky Way, tianhe 1 no. 2, the sunway supercomputers, and gradually formed the tianhe (national defense science and technology university), great power, dawn (dawn) three series of supercomputers.

At the same time, the domestic system is software and hardware system of localization in steady progress - tianhe 2 hardware system in addition to the high speed computing system adopted the Intel E5 and xeon PHI, high-speed interconnection communication network system, storage system (I/O nodes and the I/O storage management), maintenance, monitoring and control system, power system, cooling system and structure the basic implementation such as assembly design localization, in terms of software system, operating system, compiling system, parallel program development environment, most realize localization of scientific computing visualization system. While the sunway super calculate achieved in addition to the cooling system of the software and hardware of the national production.

More importantly, these supercomputers is not, as some people say that the fight for the world's first name, but the real used in ballistic calculation, nuclear physics research, climate, meteorology, Marine environment, numerical wind tunnel, collision simulation, life science, such as oil geophysical exploration research field. In addition, domestic calculate/high performance computer is widely used in industrial production, weather forecast, and entertainment.

Dawn company's high performance computers in petrochina, sinopec and cnooc exploration can be seen everywhere in the room, the high performance computer for engineer with high precision production area underground structure and geological information.

"Tianhe - cool card" cloud rendering platform from the production of animation rendering cycle shortened to 1 day, 4 to 6 months on an average day as 8 anime films at the same time provide a rendering. "Avatar" cartoon rendering production has taken more than 1 year to complete, if use "tianhe ii, is only 1 month time.

As the fog phenomenon more and more attention by the people, the fog haze weather warning forecast has become another "tianhe no.1" task. National center for super calculate tianjin has developed the fog haze automated real-time early warning and forecasting system, is in hebei baoding as a pilot, real-time forecast haze days, five days in the future national haze over the next three years to gradually build early warning and forecasting system. In addition, the tianhe super calculate can back to the earth's climate change, "number one" of the Milky Way can be used to simulate the change of 2000 years ago, "tianhe ii can be simulated and 5000 years ago and beyond.

Zhongke dawn and atmospheric physics and other units to develop numerical simulation device "" earth prototype system, then fill the blank for the practice of big data platform in the earth system.

, so to speak, a supercomputer for China's national defense, scientific research, industry, economy has made a great contribution to many aspects, such as!

History is any guide, technology blockade is actually a good thing, in the field of information technology, this period of history since the founding of can show "who can buy it is difficult to make, all can't buy to come". As long as the western technology blockade, comprador domestic and foreign is impossible "exchanging market with technology", "buy, buy than rent" to kill its own technology.

So, China's manpower and financial resources to put all into independent technology research and development, with China the world's most complete industrial sector and a solid industrial foundation, combined with China's huge market and abundant resources, can do what "" blockade, solve, so in the long run, technology blockade is a good thing.

The cores of China and the United States

In a few years ago, when the first generation of Intel xeon PHI listed, there is no similar domestic product, but the godson, ShenWei, ascended to calculate multicore chips are also far from rival Intel:

Fit 1500 cores, 40 nm process, 1.8 G frequency, maximum power 65 w, 144 G double precision floating point;

ShenWei 1600 cores, process 65 nm, 1.1 G frequency, maximum power 70 w, 140 G double precision floating point;

ShenWei 1610 cores, 40 nm process, 1.6 G frequency, maximum power consumption 50 w, double precision floating point 200 G;

The godson 3 b1000, 8 cores, process 65 nm, 1 g frequency, maximum power 65 w, 128 g double precision floating point;

The godson 3 b1500, 8 cores, 32 nm process, 1.2 G frequency, maximum power consumption of 40 w, double precision floating point 192 G.

Even the best ShenWei 1610 theory of double precision floating-point peak is only 200 g, and the first generation of Intel xeon PHI peak theory of double precision floating point up to 1 t, ShenWei 1610 is five times. It is, therefore, tianhe only helpless choice 2 Intel xeon PHI as the accelerator.

Time flies, over time, after several years of service, increasing the strength of the domestic IC design unit, not only can take out xeon PHI calculation card replacement products, and not on performance. Independent research and development of national defense science and technology university of matrix theory of 2000 double precision floating-point peak of 2.4 T, power consumption is 200 w, the theory of double precision floating-point peak reached 80% of the second generation of strong PHI, performance power stronger than slightly better than that of the second generation to PHI.

If 2000 is GSDSP matrix, rather than the nuclear accelerators, also less on performance of the second generation xeon PHI, so high performance integrated circuit design center in Shanghai to the United States all domestic nuclear chip is banned to PHI calculation card is the most powerful back, especially on the design concept, all domestic nuclear accelerators are very advanced.

With heterogeneous computing can get better performance than the power consumption and peak performance, is currently under construction in China and the United States 100 p is calculated using heterogeneous basic, also is a compute node consists of a CPU + accelerator, such as tianhe 2 a on a computing nodes by two E5 (in the future may be replaced with independent research and development of national defense science and technology university of 64 nuclear server chips "Mars") and three matrix of 2000. Accelerator can be K80 GPGPU, also can be a matrix 2000 such GPDSP, could also be an Intel xeon PHI, all domestic nuclear chip of the accelerator.

Although heterogeneous computing has many advantages, but also can bring a lot of shortcomings, such as generality is not good enough, efficiency is not high, programming more troublesome, on the other hand, the tianhe 2 and the super calculate Stampede of Intel xeon PHI calculation is not Shared memory CARDS and E5, so need the programmer explicitly copy, causing the performance loss. And homebred the nuclear chip through the innovation of design concept, greatly reduce the negative effects of the above aspects, even completely avoids the performance loss in some way.

Therefore, the author thinks that, although domestic the core with 28 nm process, in the process as the second generation of xeon PHI of 14 nm. But relying on the advanced design concept, relative to the second generation of Intel xeon PHI will have certain advantages. And double precision floating point arithmetic peak as high as 3 t performance index, makes it not as the second generation of Intel xeon PHI.

Chip manufacturing level rising in China, and is expected to shorten and the chip manufacturing level unceasingly, if Intel can't on supercomputers chip design concept has a revolutionary improvement, then by domestic beyond all the chips of the next generation of products is only a matter of time.