Saturday, November 14, 2020

The global quantum race accelerates again: Behind the tens of billions of dollars market, the hegemony of players | Annual Industry Research

 Moore's Law is approaching the upper limit, quantum computing has become an important breakthrough method

Away from the magnificent 2020, we are about to usher in a turbulent new decade full of hope and unknown. The dual promotion of technology and capital is the main theme of this decade, and it is likely to become the main thrust of innovation in the next decade. Therefore, approaching the end of 2020, we launched the "Annual Industry Research" series, and selected the most concerned areas for systematic review. These industries are either rewriting the current new economic landscape, or are likely to reshape the future business or even the international landscape, or are of interest to readers, or are having a huge social impact. We also hope to use this method to "expand the boundary infinitely" with my readers, and "see the future first" together.

This article is one in this series. We chose the industry of quantum computing. In 2020, quantum computing will continue to achieve technological breakthroughs. General Secretary Xi Jinping's speech in the collective study of the Political Bureau of the Central Committee on October 16 also brought more people's attention to quantum computing. In fact, important players in global quantum computing in 2020 will continue to achieve technological breakthroughs. In the future, this is not only the key to breaking through Moore's Law, but also an important variable affecting the international situation.


In 2020, quantum computing will continue to achieve technological breakthroughs. General Secretary Xi Jinping's speech in the collective study of the Political Bureau of the Central Committee on October 16 also brought more people's attention to quantum computing. 

In fact, important players in global quantum computing in 2020 will continue to achieve technological breakthroughs. In the future, this is not only the key to breaking Moore's Law, but also an important variable affecting the international situation.

This article will sort out the quantum computing industry from the following aspects: the history of quantum computing, technical principles and scientific research difficulties, main technical paths, application directions, comparison of development levels between China and foreign countries, inventory of players on the same track at home and abroad, etc.

1. Quantum computing is inevitable to break through Moore's Law

According to Moore's Law, the number of components that can be contained on an integrated circuit approximately doubles every 18 months, and the computing performance of the computer also doubles. At present, the integrated circuit manufacturing process is in the 14nm and 10nm technology generation stage of mass production, and the smaller size technology generation of 7nm and 5nm is in the research and development stage, which is about to reach the physical limit of control electronics. Because when a single transistor shrinks to only one or a few electrons, a single-electron transistor will appear, and the quantum tunneling effect will affect the normal operation of electronic components.

The unit of storing information in a classic computer is a binary bit (Bit), that is, a bit represents "0" or "1". The unit of information storage in a quantum computer is a qubit. In addition to "0" and "1", qubits can also represent various combinations of "0" and "1". This makes quantum computers far exceed the computing power of classical computers (what we know as "quantum hegemony" refers to quantum computing devices that exceed the computing power of all classical computers in specific test cases). Therefore, to break through the bottleneck of Moore's Law, quantum computing is a very promising choice.

2. The history of quantum computing

Since Feynman proposed the concept of quantum computer in 1981, quantum computing has entered the stage of theoretical research. However, most of the more milestone progress afterwards are in the direction of quantum algorithms. The Detusch-Jozsa algorithm clearly demonstrated the advantages of quantum computing for the first time, but the algorithm is more scientific than practical. Later, the practical algorithms proposed by people are mainly divided into two categories, namely the prime factor decomposition algorithm and the unordered search algorithm, represented by the Shor algorithm and the Grover algorithm respectively. In 1994, Shor proposed the quantum algorithm for decomposition of large numbers, exponential speeding up the decomposition of large numbers, in 1996 Grover proposed a quantum search algorithm, which accelerated the search of disordered databases by the square root In 1998, researchers from IBM, Oxford, Berkeley, Stanford, and the Massachusetts Institute of Technology produced a 2-bit computing system.

Since D-wave realized the first commercial quantum computer in history in 2007, quantum computing has entered a period of commercial explosion. In 2017, IBM announced the successful development of a 50-bit processor prototype, and "quantum hegemony" entered a critical period of contention. In 2018, Intel and Google reached 49 bits and 50 bits one after another. In 2019, Google used a 53-qubit quantum computer to complete the world's most powerful supercomputing Summit's 10,000-year computing experiment in 3 minutes, and announced that Google has achieved "quantum hegemony." In the second half of 2020, quantum computing giants are vying to release the latest scientific research results and products, including Honeywell's 10-bit QV128 ion trap quantum computer and IonQ's 32-bit 4 million ion trap quantum computer.

(Image source of quantum computing events: 36 krypton sorted according to public information)

3. Principles and difficulties of quantum computing

The unit of storing information in a classic computer is a binary bit (Bit), that is, a bit represents "0" or "1". The unit of information storage in a quantum computer is a qubit, which has two special properties: Superposition and Entang lem ent.

In addition to expressing "0" and "1", qubits can also express various combinations of "0" and "1", which is superposition. Because of superposition, the quantum computer can calculate all the results of this state at once. And this state will collapse to a certain value of "0" or "1" when observed. Therefore, although quantum computing has such a huge advantage of invisible parallel computing, we can only get one result at a time. Therefore, in order to give full play to the advantages of quantum computing, we need to design corresponding algorithms to take advantage of this invisible parallel computing advantage of quantum computing.

The nature of entanglement is reflected in the fact that a researcher prepares a pair of entangled qubits. When in the state of a single qubit, changing the state of one of the qubits will instantly change the other, even if they are far apart in the universe. This property makes the computing power of the computer increase exponentially when qubits are added to the quantum computer.

The realization of quantum computers requires quantum preparation and control capabilities. However, qubits are very fragile. Interaction with the environment will cause the quantum behavior to attenuate or even disappear. The slightest vibration or temperature change may weaken the computing power of the computer and cause calculation errors. This is also the qubit experiment that needs to be performed at ultra-low temperature or under vacuum s reason. Adding more qubits will help, but it may take thousands of qubits to ensure a highly reliable logical bit. However, adding one qubit will increase the difficulty of the experiment exponentially.

At present, various experimental platforms are far away from the tens of thousands of qubits required for practical quantum computing, and there is still room for significant improvement in the fidelity of quantum logic gates and the threshold of fault-tolerant quantum computing.

At the same time, not all calculation processes, quantum calculations have advantages over classical calculations, such as ordinary addition, subtraction, multiplication and division operations, there is no essential difference between quantum computers and classical calculations. Quantum computing has advantages only on certain problems.

Fourth, the main technical path of quantum computing

At present, there are several main technological paths for quantum computing: superconductivity, ion traps, silicon quantum dots, topology, photons, neutral atoms, molecular spins, valley qubits, etc.

Superconducting quantum computing is a quantum computing scheme based on superconducting circuits. The core device is a superconducting Josephson junction (Josephson Junction). The system requires ultra-low temperature and can use existing semiconductor micromachining processes to make microwave electronic devices. Superconducting quantum circuits are highly compatible with existing integrated circuit systems in terms of design, preparation, and measurement, and are very flexible in the design and control of the energy level and coupling of qubits. It is currently the most promising for general quantum computing. One of the realization methods.

Ion traps are also a relatively leading method of quantum computing. The ion trap does not require the extremely low temperature environment of superconducting and semiconductor systems. At the same time, the ion trap uses visible light photons, and there is no interaction between single photons so that photons do not interfere with each other, so it is more suitable for long-distance quantum communication. The disadvantages of ion traps include long interaction time, complicated experimental methods, and difficulty in integration. Current technical difficulties include how to trap a large number of ions in a small area and how to control high-precision and high-energy lasers. Therefore, the future development of trapped ions depends on two important factors, one is the manufacturing process of microwave chips, and the other is laser control.

Silicon quantum dots, that is, semiconductor quantum chips, have good scalability and integration characteristics, and are completely based on a fairly mature traditional semiconductor process, and are also suitable for quantum computing.

Topological quantum computing is an emerging cross-discipline that has been developed in the past ten years. It includes quantum computing, topology, topological quantum field theory, and condensed matter physics with topological order. It uses topological quantum states in multi-body systems to manipulate and deposit reserve sub-information is inherently fault-tolerant. The advantages of topological quantum computing include no need for large-scale error correction, strong anti-interference ability, infinitely extending coherence time, and two-bit gate fidelity up to 100%. However, compared to superconducting, semiconducting, ion trap and other directions, the development stage of topology technology is still relatively initial.

In terms of light quantum, because it does not meet the requirements of the fourth quantum logic gate, there is no interaction between single photons so that photons do not interfere with each other. This is the reason why light quantum is suitable for long-distance quantum communication, but it is also used in quantum computing. Weakness.

(Picture source of scientific research progress of the main technology path of quantum computing: 36氪 sorted out according to public information)

As shown in the figure, there are currently the largest number of superconducting and ion trap research companies and the fastest growing. The industry leaders are Google’s 72-qubit superconducting quantum computer and IonQ’s 32-qubit QV4 million ion trap quantum computer. Semiconductivity, topology, and optical quantum directions have also made progress. Specific research and development progress of Chinese and foreign quantum computers in various fields will be compared in detail later in the article.

In addition to quantum processors, technologies such as quantum software, algorithms, and cloud platforms are also crucial.

Quantum software is still in its infancy. Because the logic of quantum computing is very different from classical computing, software programmers and application developers are required to have quantum computing thinking and engineering adaptation capabilities. Quantum software development is challenging.

In terms of quantum algorithms, the Shor algorithm and Grover algorithm proposed in the mid-1990s were milestones in the development of quantum computing, after which the subsequent development of quantum algorithms was slow. At present, the core algorithm is still limited, it only has theoretical advantages on specific problems, and cannot be applied to all problems.

In terms of quantum cloud platforms, the current access channels for quantum computers include the use of specially built quantum systems and through quantum cloud platforms. Due to the harsh environmental requirements of quantum processors, high operating conditions and high maintenance costs, currently only a few companies and scientific research institutions can independently own them. The quantum computing cloud platform that accesses quantum processors through cloud services has become an important means of quantum computing applications.

V. Application direction of quantum computing

The development of quantum science has given birth to three major areas: quantum computing, quantum communication and quantum measurement.

Compared with classical computers, quantum computers are disruptive in computing power. In the initial stage of quantum computing applications, industries that consume huge time and costs, such as biopharmaceuticals, chemicals, and energy, will occupy most of the market share. In the later stage, relying on its own direct demand for computing, technology industries that require high computing capabilities, such as search, digital security, artificial intelligence, and machine learning, gradually expand their market share and become the mainstream of quantum computing applications.

Quantum Approximate Optimization Algorithm (QAOA) and Quantum Annealing Algorithm (QAA) in quantum combinatorial optimization have been applied in manufacturing, commerce, telecommunications, smart transportation, and Internet of Vehicles. D-wave's quantum annealing algorithm is world-leading. D-Wave cooperation with the German Aerospace Center DLR quantum annealing machine to achieve the optimal allocation of flight gates; and Recruit Communications and early rice fields to achieve the optimization of ad impressions university quantum annealing processor; Associated British Telecom, University College London and University of Bristol The pattern lock is an optimizable direction in the telecommunications industry; Volkswagen deploys the traffic management system developed on the Canadian D-wave quantum computer to achieve rapid route planning.

Quantum simulation technology can be used for chemical molecule or neural network modeling, applied to pharmaceutical, biological and new materials. Financial markets with extremely high complexity and randomness may also be realized through quantum simulations. QxBranch, which has been acquired by Rigetti of the United States, is cooperating with the Commonwealth Bank of Australia to develop a quantum computing simulator, which is expected to be used in bank financial services.

The combination of quantum computing and artificial intelligence is also full of expectations. The application directions include artificial intelligence natural language and image processing, artificial neural networks, etc. In April 2020, Cambridge Quantum Computing announced the success of the natural language processing test performed on a quantum computer, marking the world's first successful case. Studies have shown that the process of human brain processing information may be related to quantum phenomena, and quantum neural networks may be more suitable for simulating the information processing process of the human brain than traditional artificial neural networks.

Quantum communication is a communication method that uses the unique coherence and entanglement phenomena of quantum mechanics and uses qubits as a carrier for information transmission. Because any measurement of the quantum system will interfere with the system, it can effectively prevent third-party intrusion and protect information from eavesdropping. Quantum Key Distribution (QKD) technology enables communication parties to generate and share a set of random and secure keys for encrypting and decrypting information, but does not transmit any substantial information.

Quantum measurement is based on the precise measurement of microscopic particle systems and their quantum states. It completes the transformation and information output of the physical quantities of the system under test. It has obvious advantages over traditional measurement technologies in terms of measurement accuracy, sensitivity and stability. It mainly includes time reference, inertial measurement, gravity measurement, magnetic field measurement and target recognition. It is widely used in basic scientific research, space exploration, biomedicine, inertial guidance, geological survey, disaster prevention and other fields. 

6. Comparison of the development degree of quantum computing between China and foreign countries

Compared with the United States and other countries, Chinese technology companies have entered the quantum computing field later, and there is still a big gap between the United States in the development of quantum processors and quantum computing applications, and quantum communications are in a leading position. In recent years, the layout has been started through cooperation with scientific research institutes or hiring well-known scientists, but many are still in the exploratory stage and have not entered industrialization, and there is still a big gap compared with the international advanced level .

The main research directions of companies such as Google, IBM, Intel, Rigetti, D-wave and China Origin Quantum are superconductivity. The 72-bit computing system released by Google in 2018 is currently the industry leader. The domestic Quanyuan Quantum launched the 6-bit superconducting quantum computer "Wuyuan" this year, which is benchmarked against IBM's 2017 5-bit quantum computer. It is expected that by the end of this year, the 2-bit semiconductor computer "Wumoto" will be launched to benchmark Intel's 2018 level. From the current progress, the company's gap with IBM in the direction of superconducting technology is about 3 years, and in the semiconductor technology path, the gap with the benchmark Intel is about 2 years.

Honeywell, IonQ, and MIT Lincoln Laboratories focus on ion trap directions. The current industry leader is the 32-bit QV400 ion trap quantum computer launched by IonQ in October this year, and the ion trap quantum computer H1 with Honeywell’s 10-bit QV128 technical index. The domestic Qike Quantum "Tian Shou No. 1" ion trap quantum computer project is expected to be completed within 2-3 years. The technical indicators can reach more than 100 controllable qubits, and the quantum volume will reach 100 million. In 2022, the development of a distributed quantum computer with ion-photon entanglement will be initiated.

Companies that focus on superconductivity, such as Intel and China Origin Quantum, also focus on silicon quantum dots, or semiconductors. In October 2020,  Silicon Quantum Computing SQC in Australia  achieved an ultra-high fidelity of 99.99% of silicon double qubits, breaking the currently announced Google Sycamore's highest record of 99.64% double qubit fidelity. Origin Quantum expects to go online at the end of this year, the 2-bit semiconductor computer "Wumoto", which will be compared to Intel's 2018 level, with a gap of about 2 years.

The main direction of Microsoft's quantum computer is topology. Recently, it has cooperated with the University of Copenhagen to launch new materials for making topological computers, which will be used to realize real topological computers. There is no quantum computing company that pays attention to this direction in China.

In terms of quantum communication, my country is in a leading position. Quantum satellite is one of the leading space science projects of the Chinese Academy of Sciences. It was led by Academician Pan Jianwei. It was established in 2011. After 5 years of research preparation, it was launched in 2016 and became the world's first satellite for quantum science experiments. In 2017, the Micius successfully realized two quantum entangled photons. After being distributed to a distance of more than 1,200 kilometers, they can still maintain their quantum entanglement state. June, CAS No. achieved using Mo one thousand kilometers without repeaters entanglement based quantum protect secret communication. National Shield Quantum mainly develops in the field of quantum communication and is now a manufacturer of quantum communication equipment and a supplier of quantum security solutions. In July of this year, Guodong Quantum (stock code: 688027) was officially listed on the Science and Technology Innovation Board, making it the first A-share listed company in the field of quantum communications in China. Qike Quantum completed the commercialization of a single-photon-based quantum key distribution system in 2019, and realized quantum information transmission based on entangled states in 2020, and is currently deploying products. 

Seven, foreign quantum computing track players

 Google

Quantum processors are the main research hotspots and core bottlenecks of quantum computing technology at this stage. The main research directions include superconductivity (currently the best and fastest way to achieve solid quantum computing), ion traps (also relatively leading implementation methods) , Photons, silicon quantum dots, neutral atoms, topology, molecular spin, valley qubits, etc. Commonly used units to measure the performance of quantum processors include the number of qubits (Qubit, Q) and quantum volume (Quantum Volume, QV). Google's main research direction in quantum processors is superconductivity. In 2018, the 72Q processor Bristlecone was launched, which was the world leader in the number of qubits at that time. In October 2019, Google used the 53Q quantum computer to complete the calculation experiment that the world's most powerful supercomputing Summit took 10,000 years to complete in 3 minutes. Google researchers announced that Google has achieved "quantum hegemony."

In terms of quantum software, Google has the open source quantum computing framework Cirq. In early 2020, using TensorFlow's original strong position in the field of machine learning, the quantum machine learning library TensorFlow Quantum was launched to seize the opportunity in the field of quantum machine learning.

In August of this year, the Google quantum research team simulated the largest chemical reaction to date on a quantum computer. The research results will completely change theoretical chemistry and improve various industries, such as medicine and industry.

 IBM

IBM's main research direction in quantum processors is superconductivity. In 2017, the 50Q processor was launched. In August of this year, the Falcon processor of 27Q.QV64 and the second version of Hummingbird processor of 65Q.QV32 were successively launched. In September of this year, IBM released a roadmap for expanding quantum technology, which shows that IBM will achieve 127Q in 2021, 433Q in 2022, and 1121Q in 2023. After that, the number of qubits will reach millions.

IBM has developed a full-stack quantum software suitable for the IBM Q simulator and owns the quantum cloud platform Q Experience.

 Intel

Intel’s main focus on quantum processors is superconductivity and silicon quantum dots. In early 2018, the 49Q superconducting quantum test chip was launched, named "Tangle Lake"; in 2019, it cooperated with Bluefors and Afore to launch a quantum low-temperature wafer probe test tool to accelerate the silicon qubit test process.

 Microsoft

Topological qubits are the main concept that supports Microsoft's quantum computers. In recent years, superconductivity, ion traps, and optical quantum computing have all made remarkable achievements, but topological quantum computing has been slow to make breakthroughs. Until September of this year, Microsoft and the University of Copenhagen produced new materials for making topological quantum computers. This is a major progress Microsoft has made in researching topological quantum computers for decades.

Microsoft has launched the quantum programming language Q# and its supporting Microsoft open source quantum development kit, and has a quantum computing cloud platform Azure Quantum.

 Amazon

At the end of 2019, Amazon announced the provision of quantum computing cloud services and officially entered the quantum field. And in mid-August this year, it announced that its Braket quantum computing cloud platform was fully listed, and users can access the back-end hardware systems of D-Wave, IonQ, and Rigetti through Braket. In addition to running quantum algorithms, customers can also use Braket to design and run hybrid algorithms.

 Honeywell

Honeywell mainly studies ion trap technology in quantum processors. Following the release of QV64 quantum computer based on ion trap technology in June this year, it broke through QV128 in October.

Rigetti

Founded in 2013, Rigetti is an American quantum computing start-up company based on superconducting technology. In December 2019, Rigetti released a 32Q quantum computer.

Quantum Cloud Service (QCS) is Rigetti's quantum first cloud computing platform, and its product Forest is the world's first full-stack programming and execution environment for quantum or classical computing.

In 2019, Rigetti acquired QxBranch, a quantum computing and data analysis software startup.

In August of this year, Rigetti completed a $79 million Series C financing. So far, Rigetti has conducted 9 rounds of financing, with a total financing amount of US$269 million and 38 investors, including Bessemer Venture Partners, Franklin Templeton and many others. A well-known international investment organization.

 IonQ

Founded in 2016, IonQ is a quantum computing start-up company based on ion trap technology in the United States. In October of this year, IonQ announced the launch of the world’s most advanced QV 4 million quantum computer.

October 2019, IonQ by Samsung incubator fund (Samsung Catalyst Fund) and Mubadal A capital (Mubadala Capital) lead investor, Google Venture, Amazon and so on with a new round of investment financing, access to $ 55 million. This round of financing brings its total financing to $77 million.

 QC Ware

Founded in 2014, QC Ware is an American quantum cloud computing platform development company. QC Ware owns the quantum cloud platform Forge, and partner companies include Google and IBM.

In October 2018, QC Ware received US$6.5 million in Series A financing, led by Citi and Goldman Sachs. The total amount of financing to date has reached 14.7 million US dollars.

 Cambridge Quantum Computing (CQC)

Cambridge Quantum Computing (CQC) was established in 2014 and is a quantum computing software startup in the UK.

In April of this year, CQC announced the success of the natural language processing test performed on a quantum computer, marking the world's first successful case. In May, a consortium led by Riverlane, a Cambridge-based quantum computing software developer, received a grant of £7.6 million (approximately RMB 69 million) from the British government for the deployment of the highly innovative quantum operating system Deltaflow.OS. Three months later, Deltaflow.OS, the first universal quantum computer system developed by CQC, was born. In August, according to the CEO of CQC, CQC cooperated with IBM to develop the world's first quantum computing application.

As of today, the total amount of CQC financing has reached 50 million US dollars. Investors include Honeywell Venture Capital.

 D-Wave

D-Wave was founded in 1999 and is headquartered in Canada.

The quantum annealing machine is a kind of quantum computer that is good at solving optimization problems. D-wave's quantum annealing computer processing capacity has reached 2000Q in 2018, and it has solved many types of applications such as manufacturing, commerce, telecommunications, smart transportation, and Internet of Vehicles. Issues such as cooperation with BMW to optimize the robot movement in the manufacturing plant, cooperation with the German Space Center DLR to achieve the optimization of the distribution of flying doors, cooperation with Recruit Communications and Waseda University to achieve the optimization of advertising display, etc. In 2019, D-Wave released the 5000Q quantum annealing computer "Avantage". In October of this year, D-Wave announced that the world's first commercial dedicated quantum computer was officially launched.

In February of this year, D-wave launched the second-generation hybrid quantum computing cloud platform Leap 2.

As of today, D-Wave has raised US$210 million in financing.

In October of this year, Canada established a Quantum Industry Canada (QIC) composed of 24 quantum technology companies. Members include D-Wave Systems, a pioneer in the quantum computing industry, software developer 1Qbit, and optical quantum computer manufacturer Xanadu Quantum. Technologies, software manufacturer zapata computing , quantum security product solution provider ISARA, etc.

Silicon Quantum Computing (SQC)

Silicon Quantum Computing (SQC) was established in 2017 and is an Australian quantum computing start-up company.

SQC focuses on the silicon quantum direction in quantum processors. In October this year, it achieved an ultra-high fidelity of 99.99% of silicon atom double qubits, breaking the currently announced Google Sycamore's highest record of 99.64% double qubit fidelity.

In October of this year, John Martinis, the former head of quantum computing at Google, officially joined SQC. Martinis established Google's Quantum Hardware Group in 2014 and led the group to apply low-temperature superconducting technology to achieve "quantum hegemony" in 2019.

In December 2019, Silex Systems launched a silicon concentration project to commercialize its high-purity "zero-spin silicon". SQC and Silex signed a product offtake agreement, SQC will pay 300,000 US dollars every three years as an advance payment for the purchase of the material in the future. At the same time, SQC signed a subscription agreement to fully subscribe for Silex's common stock for USD 900,000.

So far, SQC's total financing has reached 66 million US dollars.

8. Domestic quantum computing track players

Alibaba

Alibaba is the earliest enterprise that started quantum research in China. In 2015, it began to deploy quantum computing and jointly established a laboratory with the Chinese Academy of Sciences.

At the Shenzhen Yunqi Conference in March 2017, Alibaba Cloud announced the world's first case of quantum encrypted communication on the cloud. In May , the world's first optical quantum computer that surpassed the early classical computers was born in collaboration with the University of Science and Technology of China, the Chinese Academy of Sciences- Alibaba Quantum Computing Laboratory, Zhejiang University, and the Institute of Physics of the Chinese Academy of Sciences. In September, Alibaba founded the Dharma Institute , a cutting-edge and basic scientific research institution , with quantum computing as one of its core research directions. The head of the quantum laboratory is Shi Yaoyun, a former University of Michigan professor. In the same year, the quantum cloud platform jointly built by Ali and the Chinese Academy of Sciences was launched.

In early 2018, Hungarian computer scientist Mario Szegedy joined Alibaba Dharma Academy. In the same year, the quantum circuit simulator "Tai Zhang" developed in the laboratory successfully simulated the 81-bit 40-layer Google random quantum circuit as the benchmark in the world.

In September 2019, the laboratory completed the research and development of the first controllable qubit.

In March of this year, Alibaba Dharma Academy launched the Nanhu project with a total investment of about 20 billion yuan. The main research direction includes quantum computing. In June, Alibaba's innovative research plan AIR included the quantum plan for the first time.

 Tencent

Tencent began to deploy quantum science in 2017. Ge Ling, PhD in quantum computing from Oxford University, joined Tencent as the chief representative of Tencent Europe.

In 2018, Professor Zhang Shengyu, a well-known quantum theory computer scientist at the Chinese University of Hong Kong, joined Tencent and established Tencent Quantum Lab. In the same year, Tencent proposed the "ABC2.0" plan (AI, Robotics, Quantum Computing).

Tencent is conducting related research in application areas such as quantum AI, drug development and scientific computing platform (SimHub).

 Baidu

In March 2018, Baidu established the Quantum Research Institute, and Professor Duan Runyao, the founding director of the Quantum Software and Information Center of the University of Technology Sydney, served as the director. The research institute focuses on the research and development of quantum algorithms, quantum AI applications, and quantum architecture. It develops quantum computing platforms and connects with different quantum hardware systems through flexible and efficient quantum hardware interfaces, and finally outputs quantum computing in the form of cloud computing. ability.

At the 2019 Developer Conference, Baidu released the high-performance quantum pulse computing system "Quantity Pulse".

In May of this year, Baidu released the country's first quantum machine learning development tool "Paddle Quantum." In September, Baidu released a brand-new upgraded Baidu Brain 6.0, which includes the country's first cloud-native quantum The computing platform "Quantity Yifu" realizes the deep integration of quantum computing and cloud computing.

 Huawei

Weng Wenkang announced in 2018 to join the Huawei Data Center Technology Laboratory. The main research directions of the laboratory include quantum computing physics and manipulation, quantum software, quantum algorithms and applications, etc. In the same year, Huawei announced the quantum computing simulator HiQ 1.0 cloud service platform for the first time.

In 2019, HiQ was upgraded to 2.0. A single Kunlun quantum computing and simulation machine can achieve full-amplitude simulation of 40 qubits and single vibration simulation of up to 144 qubits (22 layers).

In September this year, HiQ was upgraded to 3.0.

Hon Hai Group

Foxconn's parent company established the Hon Hai Research Institute in March this year, and invited Taiwan University Physics Distinguished Professor Zhang Qingrui as the leader of the quantum computer project to start the quantum computing layout. 

TSMC

Taiwan Semiconductor Manufacturing Company ( TSMC ) is some of the top hardware manufacturers in the technology industry. In 2018, it plans to cooperate with Taiwan's Ministry of Science and Technology (MOST) to create a cloud quantum computing platform based on IBMQ.

 National shield quantum

National Shield Quantum was founded in 2009, originated from the University of Science and Technology of China, led by Pan Jianwei, Executive Vice President of the University of Science and Technology of China. National Shield Quantum mainly develops in the field of quantum communication and is now a manufacturer of quantum communication equipment and a supplier of quantum security solutions. In July of this year, Guodong Quantum (stock code: 688027) was officially listed on the Science and Technology Innovation Board, making it the first A-share listed company in the field of quantum communications in China.

Origin quantum

Origin Quantum was established in 2017, relying on the Key Laboratory of Quantum Information of the Chinese Academy of Sciences, led by Academician Guo Guangcan and Professor Guo Guoping, two leaders in the quantum computing industry. The goal of creation is the development of full-stack quantum computing, directly against foreign quantum computing companies IBM, Rigetti, etc., and its main business covers quantum chips, quantum measurement and control, quantum software, quantum clouds and future quantum applications.

In terms of quantum chips, Yuanyuan has developed the first-generation semiconductor 2-bit quantum processor Xuanwei XW B2-100, and the first-generation superconducting 6-bit Kuafu quantum processor KF C6-130. In the field of quantum measurement and control, Origin has successfully developed the first domestic quantum computer control system-OriginQ Quantum AIO, the first-generation quantum measurement and control integrated machine. In terms of quantum software, Yuanyuan developed the first set of quantum language standards in China, QRunes, and developed the first domestic quantum programming framework QPanda (a composite architecture of quantum language and compiler), the first domestic quantum computing application framework pyQPanda, and the first domestic quantum Program development plug-in Qurator-VSCode. In the field of quantum cloud, Origin has developed a free experience platform for quantum computing based on 32-bit quantum virtual machines. The original superconducting quantum computing cloud platform, which was launched in September 2020, is connected to the 6-bit superconducting quantum chip "Kuafu" based on real physical systems. The quantum computer assembled from this 6-bit quantum chip is named "Wuyuan" (another quantum computer based on semiconductor technology that has not yet been released is named "Wumoto").

Recently, in the global quantum computing technology invention patent rankings released by the intellectual property industry media IPRdaily and the incoPat Innovation Index Research Center, Origin Quantum ranked 7th with 77 patents, and the top 6 were IBM, D-Wave, Google, Microsoft, Northrop Grumman, Intel.

 Guoyi Quantum

Guoyi Quantum originated from the Key Laboratory of Micromagnetic Resonance, Chinese Academy of Sciences, University of Science and Technology of China, led by Academician Du Jiangfeng. With quantum precision measurement as its core technology, Guoyi Quantum provides high-end equipment and services.

In April 2019, Guoyi Quantum launched the diamond quantum computing teaching machine, which is mainly used for quantum computing experiment teaching. Its design qubit number is 2 qubits and can work at room temperature.

Qike Quantum

The core personnel of Qike Quantum led the development of the world's first commercialized quantum communication system Navajo in 2003, and has continuously developed a total of four generations of commercialized quantum communication systems. At present, while operating the domestic market, the company is also actively expanding overseas markets. In terms of quantum computing, Qike Quantum led the research and development of the world's first quantum computing measurement and control system in 2015. It is expected to complete the "Tian Shou No. 1" ion trap scalable distributed quantum computer within 2 to 3 years, with up to 100 controllable qubits.

 Ask the sky quantum

In September of this year, Wentian Quantum released the quantum education cloud platform, which is mainly for people interested in quantum, providing an ecological community of quantum learning, and related solutions, and providing virtual simulation software Qsim and Qlab cloud platforms.

 Measuring Spin Technology

In October of this year, Liangxuan Technology released the latest generation of universal quantum cloud platform "Taurus". The platform has been connected to a 2-qubit and a 4-qubit nuclear magnetic resonance quantum computer. The connected nuclear magnetic resonance quantum computer is named "Gemini", which is very small and can run at room temperature.

Chinese Academy of Sciences

In 2017, Alibaba and the Chinese Academy of Sciences launched a quantum computing cloud platform.

In December 2019, the Chinese Academy of Sciences released isQ, China's first quantum programming platform.

Quantum satellite is one of the leading space science projects of the Chinese Academy of Sciences. It was led by Academician Pan Jianwei. The project was established in 2011. After five years of research preparation, it was launched in 2016 and became the world's first satellite for quantum science experiments. In 2017, the Micius successfully realized two quantum entangled photons. After being distributed to a distance of more than 1,200 kilometers, they can still maintain their quantum entangled state. In June of this year, the Chinese Academy of Sciences used the Mozi to implement entanglement-based, unrelayed thousand kilometers quantum secure communication.

 University of Science and Technology of China

In 2017, the team of Academician Guo Guangcan of the University of Science and Technology of China, Chuanfeng Li and Guoyong Xiang's research group cooperated with Fudan University, Beijing Institute of Technology, and Nanjing University of Posts and Telecommunications to achieve the world's most efficient quantum state tomography measurement, which was published in the international authoritative journal Nature Communications. In the same year, the Pan Jianwei team of the University of Science and Technology of China successfully realized topological data analysis on the optical quantum processor.

In 2019, the University of Science and Technology of China developed 24 superconducting qubit processors.

In March of this year, Essence Quantum and the University of Science and Technology of China made important progress in the coherent manipulation of phonon modes of nano-harmonic oscillators.

 Tsinghua University

In 2011, Yao Qizhi founded the Center for Quantum Information (CQI) of Tsinghua University.

In 2017, Jin Qihuan, Associate Professor of the Quantum Information Center of the Institute of Interdisciplinary Information, Tsinghua University, led the ion trap quantum computing research group to achieve single-qubit storage with a coherence time of more than 10 minutes. The results have been published in "Nature-Photonics". In the same year, Tsinghua University released the world's first nuclear magnetic resonance quantum computing cloud platform.

In April 2018, Professor Duan Luming's research group from the Institute of Interdisciplinary Information of Tsinghua University published a study titled "Experimental Implementation of Entanglement between 25 Independently Controlled Quantum Interfaces" in the journal Science Development. paper. In the same year, Shing-Tung Yau Mathematical Sciences Center and assistant professor of Jinlong Dyatlo Semyon v teach teach cooperation paper "double-face semi-classical songs with a full measure of support set" (Semiclassical measures on hyperbolic surfaces have full support) in the top international mathematics journal "Acta Mathematica" ( "Journal of Mathematics") published online. The results of this paper are of great significance for understanding quantum chaotic systems.

 Beijing Institute of Quantum Information Science

In December 2017, the Beijing Municipal Government and the Chinese Academy of Sciences, the Academy of Military Sciences, Peking University, Tsinghua University, Beijing University of Aeronautics and Astronautics and other units jointly established the Beijing Institute of Quantum Information Science. Xue Qikun, academician of the Chinese Academy of Sciences and vice president of Tsinghua University, served as the dean.

 Shanghai Quantum Center

In June 2019, the Chinese Academy of Sciences and the Shanghai Municipal People's Government jointly established the Shanghai Quantum Science Research Center.

 Nanjing University

In July 2019, Nanjing University developed the world's first unmanned quantum communication network, using drones as a transit station for quantum transmission. On June 26, Wang Shuming, Zhang Lijian, Wang Zhenlin, Zhu Shining and other collaborators of the State Key Laboratory of Solid State Microstructure Physics of NTU and the Hong Kong Polytechnic University and the University of Science and Technology of China made research results based on " High-dimensional quantum entanglement and multiphoton light source based on meta- lens array" The topic was published in the latest issue of "Science" magazine.

 Nanjing Xianyangjian Quantum Computer Research Institute

Nanjing Xianyangjian Quantum Computer Research Institute was established in 2018. It is a state-owned state-owned enterprise jointly invested and established by the State Key Laboratory of Solid Microstructure Physics and the State Key Laboratory of Computer Software New Technology of Nanjing University. Capital holdings of research institutions.

 Zhejiang University

In 2018, Zhejiang University launched the Quantum Project. In August 2019, Zhejiang University, the Institute of Physics of the Chinese Academy of Sciences, the Institute of Automation of the Chinese Academy of Sciences, and the Beijing Computing Science Research Center jointly developed a quantum chip with 20 superconducting qubits. The research results were published in the journal Science.

 Institute of Quantum Science and Engineering, Southern University of Science and Technology

The Institute of Quantum Science and Engineering of Southern University of Science and Technology was established in 2017 and relies on Southern University of Science and Technology. The first dean of the Institute was Professor Yu Dapeng, an academician of the Chinese Academy of Sciences.

In addition, more and more universities, research institutes and enterprises such as Yaguang Technology, China Quantum , HKUST Guochuang, and Haofeng Technology have begun to enter this track to deploy quantum computing or conduct related scientific research and commercial activities.

Yitu listed, the AI ​​four little dragons headed towards the "three forks"

Computer vision is approaching its limit, where are the four AI dragons going.

Following Yitu Technology's application for IPO, the AI ​​four dragons officially parted ways.

November 4 evening, the Shanghai Stock Exchange, information display, in accordance with plans Technology Branch board submitted a prospectus. Like No.9 Co., Ltd., Yitu Technology uses CDR (Chinese Depository Receipts) to issue. It is expected to issue 291 million CDRs. It plans to raise 7.505 billion yuan. The funds raised are mainly used for new generation artificial intelligence IP and high-performance SoC Chip project. If the inquiry and deliberation go smoothly, Yitu Technology is expected to become the first listed AI company among the AI ​​Four Dragons, and it will also become another listed AI chip startup after the Cambrian.

In fact, Yitu Technology’s initial positioning or betting on the track is not an AI chip company.

In 2012, Zhu Long, a PhD in statistics from the University of California, Los Angeles, who returned from the United States to focus on computer vision, and Lin Chenxi, a senior expert in Alibaba Cloud Computing, jointly founded YITU Technology. Since 2013, Yitu Technology has received eight rounds of financing. Prospectus shows, according to Figure controlling shareholder Yitu Holdings, Long Zhu Lin Chenxi signed with concerted action, the shares were held YituHoldings 63.316%, 36.684%, the second largest shareholder is the founder of Ma Yunfeng fund, holding 10.8124%.

As soon as he debuted, he had high-quality capital blessings, the best time for the third wave of AI to rise, and the best direction CV (computer vision) in the AI ​​field where he is located. Yitu Technology has quickly become an AI start-up alongside Megvii, SenseTime, and Yuncong, and is known as the AI ​​Four Dragons (or CV Four Dragons).

However, after two years of vigorous advancement in the field of computer vision, limited by the algorithm bottleneck, homogeneity, and difficulty in landing the scene, it began to lack stamina. Since 2018, there have been endless speculations about the profitability of the AI ​​Four Dragons and the strategic route of financing while investing. However, because the AI ​​Four Little Dragons are all start-ups, there is no public financial data disclosure, and they eventually fell apart.

It wasn't until 2019 that Megvii submitted the prospectus to the Hong Kong Stock Exchange, and the outside world really saw the "hidden corner" of the survival of the AI ​​four dragons. Similar to Megvii, Yitu Technology's growth trajectory also includes high R&D investment, high growth, high losses, and "leapfrog" strategic upgrades. But betting on AI chips, Yitu has gone in a different direction from the AI ​​Four Dragons back then.

"From soft to hard", speed up the track change

Unlike last year's rumor that Yitu has achieved balance of payments, losses are its "normal".

The prospectus disclosed that Yitu's revenue doubled from 2017 to the first half of 2020, from RMB 68,718,900 in 2017 to RMB 304 million in 2018, and RMB 717 million in 2019. The losses are also deepening, with losses of 1.168 billion, 1.168 billion, 3.647 billion, and 1.303 billion, respectively, and a total loss of 7.28 billion in three and a half years.

In contrast, Megvii's accumulated losses from 2016 to the first half of 2019 were 9.65 billion. The loss was mainly due to changes in the fair value of Megvii's preferred stock and continued R&D investment. Yitu’s losses also came from high R&D investment and changes in the fair value of preferred stocks.

In three and a half years, Yitu's R&D expenses accounted for more than 90% of its operating income, and it exceeded 100% in 2017 and the first half of 2019. The three-and-a-half years' worth of preferred stock fair value losses totaled 5.083 billion, accounting for 70% of the total loss.

Shen Meng, executive director of Chanson Capital, explained the so-called change in the fair value of preferred shares to Geekpark (ID: geekpark), "The change in fair value of preferred shares is just an accounting treatment, and the impact on the company’s net profit is actually a non-cash Subjects have essentially no impact on the company’s continuing operations and cash flow.” He emphasized that to a large extent, it is a digital game. In order to make business losses look less lonely and conspicuous, accounting losses are put together. , It doesn't look so dazzling.

In other words, whether it is despising or Yitu is still a huge loss.

At the same time, Yitu began to bet heavily on the new track. In 2017, Zhu Long saw the mismatch between complex algorithms and computing power, proposed "algorithms as chips" and invested in AI chip start-up Shanghai Yizhi Electronic Technology. Two years later, YITU cloud customized chip "Quest" came out, especially for video applications.

In the prospectus, Yitu positions itself as an AI company that focuses on artificial intelligence chip technology and algorithm technology, and develops and sells artificial intelligence solutions including artificial intelligence computing hardware and software. That is, it has AI chip design capabilities and AI algorithm capabilities.

Yitu’s competitors in the same industry have also changed. The first category is giant companies such as Google and Huawei; the second category is AI chip design companies such as Nvidia and Cambrian; the third category is Hikvision and iFLYTEK , SenseTime, and Kuang are regarded as smart public services and smart business companies that provide AI industry solutions.

With the addition of the chip track, Yitu has not only completed computer vision, voice, and chip "jumps", but also expanded its benchmark targets from the AI ​​Four Dragons to giant companies such as Google and Nvidia. At the 2019 "Quest" chip conference, Yitu even labeled "No chip beyond NVIDIA is meaningless". Yitu officially "leaves" the AI ​​Four Dragons camp, moving towards differentiation.

Yitu CEO Zhu Long explained at the 2019 search conference|Yitu Technology

Under the "Sanchakou", the "difficult to realize" is difficult to crack

When looking back at the influx of the AI ​​Four Dragons into the computer vision face recognition track in 2013, there were differences in the same business line distribution, but the differences were not obvious.

For example, SenseTime is similar to Megvii. Its business lines are concentrated in the fields of security, mobile phones, and online finance. SenseTime has an autonomous driving business line; Yitu and Yuncong are mainly security and offline ATM finance. One more line of medical business. On the whole, the AI ​​four dragons have the same thing in that the security business accounts for the largest proportion of the total business.

However, the security field is still firmly controlled by the traditional video surveillance giants Hikvision, Dahua Co. , Ltd. , and Univision Technology, which have a market share of nearly 70% to 80%. The AI ​​four dragons can only be in the remaining market space. Get together and compete. In addition, in recent years, traditional video surveillance companies including Hikvision and Univision Technology have also invested a lot of resources and funds to actively deploy AI computer vision technology.

An industry insider who did not want to be named revealed to Geek Park that so far, in addition to price factors in the security field, hardware capabilities are more important. For example, camera sensors involve the upstream and downstream of the industry chain, and it is very difficult to integrate well. The AI ​​Four Dragons can only cut in from software capabilities, and software is difficult to make a profit. Of course, except for certain situations, such as high-precision identification in the financial and medical fields, the threshold is very high. But embarrassingly, the financial and medical landing scenes seem to be many, but they are not yet mature. The high R&D investment is difficult to land and recover the cost.

Even a giant such as Huawei is quite frustrated in the face of the large-scale market established by the security "leading" company Hikvision. At the beginning of this year, Huawei’s security business line was officially renamed "Huawei HoloSens Machine Vision." Except for the expansion of its business to industrial vision, robotics, and ADAS (advanced driver assistance systems), the unwillingness to run out of energy on the Red Sea may be root cause.

Driven by the same reasons, in 2019, the AI ​​Four Dragons will accelerate their differentiated development.

At the beginning of last year, Megvii cut into the robot AIoT field from a single point of computer vision. It not only released the software intelligent robot operating system Hetu, but also released a variety of robot hardware products to deploy industry, warehousing and logistics. Yitu has gone from computer vision, speech technology and natural language processing (NLP) to cloud AI chips. SenseTime and Cloud shifted to the field of smart cities, and SenseTime entered the cloud and edge, and launched the Ark operating system, the urban vision center. Cloud has been upgraded from a human-machine collaborative operating system and solution provider, starting from the industry, combining software and hardware to create a human-machine interface.

The AI ​​Four Little Dragons have moved towards the "three forks" in terms of business strategy and play style.

On the one hand, there is not much growth potential and is approaching the limit of the computer vision face recognition market, on the other hand, the more difficult robot, chip hardware market, and the smart city operating system software market. Whether it is the rumors of the AI ​​Four Little Dragons listing in the past two years or the "transformation" of the business strategy, after all, it is to solve the problem of insufficient corporate cash flow after the capital market returns to calm.

The AI ​​four dragons have fully expanded from single-point business to multi-threaded business. Does it increase market valuation, ease the anxiety of funds, or further deepen the business gap, listing may be the "only solution" to the current dilemma.

Friday, November 13, 2020

AI performance benchmarks have since had "Chinese standards". Nvidia and Google can try this set of computing power rolls

 Domestic ``AI Test Paper''-AIPerf

In the case of computing power, an AI performance benchmark called MLPerf has often jumped into people's attention in recent years.

In order to use this standard to prove the strength, the performance of big "computing power" companies such as Nvidia and Google can be said to have earned enough attention.

As early as December 2018, when MLPerf was first released, NVIDIA based on its Tesla V100, achieved excellent results in six tests including image classification, object segmentation, and recommendation system, and won the best of the audience.

Since then, Nvidia has repeatedly brushed the list. In the latest performance test not long ago, Nvidia broke eight AI performance records with the A100 GPU.


Google did not show any weakness, with 4096 TPU V3 reducing the BERT training time to 23 seconds.

In this regard, Jeff Dean, the head of Google AI, also posted on social platforms:

I am very happy to see the results of MLPerf 0.7, Google TPU has set six records in eight benchmark tests.

We need (change) a larger standard, because we can now train ResNet-50, BERT, Transformer, SSD and other models within 30 seconds.


So the question is, is the MLPerf set of "exam questions" that these "computing power" companies are chasing, really the "only standard for AI performance benchmarking"?

not necessarily.

To achieve the ideal AI or high-performance computing (HPC) benchmark, there are three challenges:

  • First of all, the benchmark workload (workload) needs to express actual issues about hardware utilization, setup costs, and calculation modes.

  • Second, it is best for benchmarking workloads to automatically adapt to machines of different sizes.

  • Finally, using simple and fewer indicators, you can measure the overall system performance of AI applications.

On the other hand, MLPerf, as Jeff Dean said, it has a fixed workload size, which may be a mistake in itself.

Because the increased computing power should be used to solve larger-scale problems instead of using less time to solve the same problems.

And benchmark tests like LINPACK cannot reflect the cross-stack performance of AI without representative workloads.

In response to the above problems, Tsinghua University, Pengcheng Laboratory, and the Institute of Computing Technology of the Chinese Academy of Sciences jointly launched a set of "Chinese AI Test Papers"-AIPerf.

Simply put, the characteristics of AIPerf are as follows:

  • It is based on the automated machine learning (AutoML) algorithm, which can realize the real-time generation of deep learning models, has adaptive scalability to machines of different sizes, and can test the effect of the system on general AI models.

  • Calculating the amount of floating-point operations through a new analytical method can quickly and accurately predict the floating-point operations required in AI tasks, so as to calculate the floating-point operations rate and use it as an evaluation score.

So, how difficult is this set of "AI test papers" in China? Scientific or not?

Please continue reading.

What does this set of "AI test papers" in China look like?

Spread out this set of "AI test papers", the full picture is as follows:

△ AIPerf benchmark work flow chart

As mentioned earlier, AIPerf is implemented based on the AutoML algorithm. In terms of framework, the researchers chose a more user-friendly AutoML framework—NNI (Neural Network Intelligence).

But on this basis, the researchers modified the NNI framework to address issues such as "AI accelerator idle" and "time-consuming model generation".

The workflow of AIPerf is as follows:

  • Access the master node through SSH, collect the information of the slave nodes, and create a SLURM configuration script.

  • Through SLURM, the master node distributes the workload in parallel and asynchronously to the slave nodes corresponding to the request and available resources.

  • After the slave nodes receive the workload, they perform architecture search and model training in parallel.

  • The CPU on the slave node searches for a new architecture according to the current historical model list (the list contains detailed model information and accuracy on the test data set), and then stores the architecture in a buffer (such as a network file system) for later training.

  • The AI ​​accelerator on the slave node loads the "candidate architecture" and "data", uses data parallelism to train with HPO, and stores the results in the historical model list.

  • Once the conditions are met (such as reaching the user-defined time), the run will terminate. The final result is calculated based on the recorded indicators and then reported.

After completing this set of "AI test papers", how should the scores obtained be measured and ranked?

We know that FLOPS is currently the most commonly used performance indicator to reflect the overall computing power of high-performance computing.

In this set of "test papers", researchers still use FLOPS as the main indicator to directly describe the computing power of AI accelerators.

In AIPerf, the floating-point operation rate is treated as a mathematical problem to solve. By decomposing the deep neural network and analyzing the calculation amount of each part, the calculation amount of floating-point number is obtained.

Combined with the task running time, the floating-point operation rate can be obtained and used as the benchmark score.

Once the theory is in place, the experiment must keep up.

The hardware specifications are as follows:

The details of the assessment environment are as follows:

Finally, announce the performance results!

Researchers ran the AIPerf benchmark test on machines of various sizes, mainly evaluating two aspects, namely stability and scalability.

From 10 nodes to 50 nodes, there are up to 400 GPUs. All intermediate results, including the generated architecture, hyperparameter configuration, precision and time stamp at each point in time, are recorded in the log file.

The following figure shows the changes over time of the "benchmark score" and "standard score" (both in FLOPS) evaluated by machines of different sizes.

The results show that the AIPerf benchmark test has robustness and linear scalability.

Next, is the relevant evaluation of GPU and its memory utilization under different scale machines.

It can be seen from the figure that the overall calculation and memory utilization of the AI ​​training card is very high (both are greater than 90%). In the transition phase between different models, due to data loading and calculation graph compilation, the utilization rate will decrease.

Why is this "examination paper" produced?

After "viewing the test paper", there is a question to think about:

Why is there a set of AI benchmark tests called AIPerf?

This issue needs to be viewed from the outside to the inside.

First of all, from the appearance, similar to MLPerf and LINPACK benchmark test programs, there are some loopholes and problems in itself:

  • Either the size of the workload is fixed, and the increase in computing power should be used to solve larger-scale problems, which limits scalability.

  • Or in the absence of representative workloads, the system's cross-stack computing performance for AI cannot be reflected.

Although such evaluation standards are of certain value and significance at present, the objective deficiencies cannot be ignored.

After all, in the current environment of rapid development of artificial intelligence, computing power is particularly important, and a complete and more scientific "benchmark" will help the development of computing power.

From this point of view, "benchmark test" and "computing power" are more like a pair of force and reaction force.

Secondly, from a deep perspective, it is very necessary to develop computing power.

For high-performance computing, the "TOP500" list was born as early as 1993. From the initial dominance of the United States and Japan to the rise of China's computing power, it is not difficult to see the country's investment in this construction.

The reason is simple. High-performance computing plays a vital role in the development of aerospace, petroleum exploration, water conservancy projects, and emerging high-tech industries in various countries.

However, with the rise of AI, the traditional "solving method" of high-performance computing has changed-AI+HPC is the future development trend of computing power.

In recent years, the TOP500 list can reflect this:

  • The first ARM architecture HPC to top the list is based on Fujitsu's 48/52 core A64FX ARM.

  • SUMMIT, ranked second, uses IBM Power+NVIDIA V100.

  • ...

Nearly 30% of the systems on the list have accelerator cards/coprocessors. In other words, more and more systems are equipped with a large number of low-precision arithmetic logic units to support artificial intelligence computing power requirements.

In our country, more and more companies have begun or have already deployed them.

Such as Huawei, wave, Lenovo, are tough to come up with their own products, such as TOP500, MLPerf etc. list in the displayed their skill.

From a practical perspective, you may think that developing computing power is of no use to ordinary people, but it is not.

It just so happens that the "Double 11" shopping mall every year is coming, and behind every e-commerce platform, there is a powerful recommendation system, which is the "guess you like" function that users often see.

Whether the recommendation is accurate or fast depends to a large extent on the strength of the AI ​​computing power.

Moreover, the annual turnover of hundreds of billions of yuan can ensure the success of timely payments, and AI computing power is also indispensable.

...

Finally, back to the original question:

What will be the performance of this set of "AI test papers" issued by China, namely ALPerf, Nvidia, Google and other established computing power companies?

Introducing AIrbq.com: Your Go-To Source for the Latest in AI Technology News

Introducing AIrbq.com: Your Go-To Source for the Latest in AI Technology News AI technology is rapidly advancing and changing the way we liv...