标签归档 南京新茶外卖论坛

How many steps to excavate RISC-V gold mine?

Author | Bao Yonggang
Edit | Wang Chuan
Sometimes the future runs faster than people think.
As Liu Cixin wrote in "Three-body", human beings are a race with "technological explosion".
The rolling wheel of the times constantly reshapes people’s cognition of technology. Even those who light a single spark can hardly imagine the future wildfire.
Moakley and Eckert, who invented the world’s first general-purpose computer in 1946, could never have predicted that the computing power once obtained by a 170-square-kilometer, 30-ton behemoth can now be easily obtained through cloud computing.
John mccarthy, who put forward the concept of artificial intelligence in 1961, can hardly imagine the scene where AI assistants are active at people’s fingertips today.
Martin cooper, who invented the mobile phone in 1990s, never imagined that the mobile communication tools he developed almost evolved into "external organs" of human beings one day, and immersed in all aspects of life.
Behind these unexpected things, there is a reason: the evolution of the underlying processor.
At the end of the 20th century, Arm CPU did not enter the PC market smoothly, but after turning to the mobile CPU market and changing its business model, it finally came to a bright future and made the mobile Internet prosperous.
At the beginning of the 21st century, x86 CPU, which flourished in the PC processor market, unexpectedly entered the server market with the advantage of cost performance and promoted the development of cloud computing.
In 2010, RISC-V, a new instruction set, was born at the University of California, Berkeley. In 2014, the first people in China who came into contact with RISC-V were attracted by the open source characteristics of RISC-V, but no one knows how RISC-V will develop in the future.
When Ali Pingtou’s team came into contact with RISC-V in 2014, they were also confused about the prospect of commercialization of this new instruction set. Brother Pingtou, who recognized the open source concept, first verified the technical capability of RISC-V and then proved the high performance of RISC-V.
In the autumn of 2022, Brother Pingtou rushed into the blue ocean of RISC-V energy-efficient CPU with Xuantie C908, and wanted to explore the edge AI market.
This is the key battle of Pingtou, and it will also be an important milestone on the way of RISC-V.
The "accident" that RISC-V will bring to the technology industry lies in the development of RISC-V in recent years.
01
Energy-efficient blue ocean,
RISC-V comes up with two killers.
In 2014, RISC-V was only active in the circle of a few technology enthusiasts. “In 2014, we were also somewhat uncertain about whether we should fully invest in RISC-V.Meng Jianyi, vice president of Pingtou Brother, said, "I am mainly worried about whether this instruction set from academia is capable enough, and I don’t know the advantages of its functions."
In fact, this is also the concern of most people who hold a wait-and-see attitude towards RISC-V. However, under the appeal of the open source concept, with the joint efforts of academic circles and the contributions of technical geeks, RISC-V was introduced into China in 2018 and began to attract attention.
In this year, Pingtou, who came into contact with RISC-V early, also took the lead in developing the first RISC-V CPU-

Xuantie E902, a CPU with low performance and low power consumption. In fact, this is the product of Pingtou Brother’s test of RISC-V. After verifying the low power consumption of RISC-V, Pingtou Brother introduced a Xuantie C910 with a main frequency performance of 2.5GHz a year later, and its ability to make high-performance products with RISC-V was also verified.

Having enough confidence in RISC-V, in 2020, Pingtou launched Xuantie C906, which has a more cost advantage. It is also the first commercial CPU in the world whose main frequency exceeds 1GHz. A year later, Brother Pingtou launched Xuantie E907 with improved performance.
marketalwaysEnergy-efficient RISC-V products at Arm Cortex A55 level are always absent.
Cortex A55 is a product launched by Arm as early as 2017. Until 2022, many products, including home appliance SoC and automobile MCU, still use Cortex A55 core. This is enough to show the market demand for energy-efficient CPU, so at the 2022 Yunqi Conference, Pingtou released the latest generation of energy-efficient CPU Xuantie C908.
Since it is a latecomer product, if you want to succeed in the market, you must have more attractive features.
"Xuantie C908 is the most energy-efficient product in Xuantie’s family now."Meng Jianwei introduced that Pingtou promoted technological innovation from the whole CPU architecture and AI energy efficiency.
While having higher performance than Arm A55 of the same class,Xuantie C908′ s higher energy efficiency and native support AI are its two killers in the energy-efficient blue ocean market.
The data shows that the main frequency performance of Xuantie C908 is up to 2.0GHz, which is 40% higher than that of Xuantie C906 and 24% higher than that of Xuantie C906. It natively supports Vector 1.0, and has more than 20% energy efficiency advantage compared with Cortex A55.

"RISC-V itself has the advantage of streamlining. To further achieve high energy efficiency under a streamlined architecture, what we can do is to design the entire hardware circuit as simplified and ingenious as possible to achieve a more extreme energy efficiency ratio." Meng Jianwei said.
It is a great challenge to realize high-performance and energy-efficient CPU. In order to improve energy efficiency, Pingtou brother has made various innovations. On the one hand, he chose a 9-stage dual-launch assembly line; On the other hand, the instruction fusion technology is adopted, that is, two instructions are turned into one for execution; Also, the energy efficiency is improved through the optimization of memory bandwidth.
Native support AI, as the killer of Xuantie C908 to achieve differentiation, also needs to consider energy efficiency. Based on the standard Vector architecture, Pingtou has made an extended design to support the new Int4 data format, which makes Xuantie C908 support AI natively and take into account energy efficiency.
Customers who use C908 will no longer need a separate NPU if their computing power demand for AI is less than 1TOPS.For this customer, it not only eliminates the trouble of choosing from dozens of NPUs in the market, but also reduces the difficulty of development.
Don’t underestimate the AI computing power of 1TOPS,This computing power is enough to meet the intelligent needs of most current smart cameras, and it can also develop the recently popular AI scanning pen.
"An emerging architecture is definitely looking for emerging markets and emerging applications. becausefrommarketMost productsThere is room for development.The link is the first to break through, and it is easier to form a demonstration of large-scale application."Meng Jianyi said.
Battery-powered products with demand for AI are just the blue ocean suitable for RISC-V energy-efficient CPU to take the lead in breaking through. The demand for intelligence in this field is becoming more and more popular, but the challenge in this field lies in ecology.
Besides differentiated technologies, Pingtou Brother will also provide a complete set of solutions, which is also different from Arm’s way, which makes it easier for users to develop and make more free choices.
02
Nugget edge AI
Four gold mines, ecology is the strongest tool.
Xuantie C908 has very clear application scenarios-intelligent interaction, multimedia terminal, AR/VR, wireless communication, etc. “Xuantie C908 should be the largest product among our high-end products."Meng Jianyi judgment.
The urgency of market demand for products can be seen from the speed of commercialization. Leifeng. com learned that,Xuantie C908 already has many early users,One of them is Jianan Technology, which has also worked in RISC-V field for many years. Based on Xuantie C908, Jianan Technology is developing a high-end stereo vision AI chip.
If we want to push our products to a wider market, we need ecological prosperity.The relationship between chips and upper ecology is like the relationship between roots and trunk branches,Only when the chip has enough capabilities can the upper software ecology prosper.
To measure a processor, we should not only look at the hardware,The ability of basic software is also indispensable, which is the consensus of the industry.
"We provide complete development tools, and at the same time we continue to invest.This year, our most important progress is that we already have a very rich support of third-party commercial tools.There are also more and more open source tools to support our processors. "Meng Jianyi said," We have cooperated with some very influential companies in the ecology, and the support of third-party commercial and open-source tools will be the characteristic of our processors. “
At the same time, the advantage of commercial and open source toolchains is that it is easier to solve the problems encountered in the development process.
For example, in the RISC-V ecosystem, debugging is not well done, and Pingtou has a better plan. Brother Pingtou provides HHB compilation tool, which can not only make good use of the AI capability of Xuantie CPU, but also support the AI IP of third parties.
On the next level, TMedia, a cross-platform video vision application framework launched by Pingtou Brother, has the characteristics of high elasticity, high energy efficiency and high scalability, and better supports customers in developing applications.
Only the software tools at the bottom of the chip are still not enough, and the mainstream operating system needs to be supported. In the low-power MCU market, the application of RISC-V does not depend on the software ecology, which is the key reason why RISC-V can ship 10 billion chips even without a mature ecology.
If you want to dig gold in the energy-efficient edge computing market, you can’t get around the problem of software ecology.This year, a major development of RISC-V is that Google Android natively supports RISC-V processors,Andriod has a very large and rich mobile ecosystem. Andriod’s native support for RISC-V adds a new space for the application of RISC-V in the mobile market.
"This is symbolic, Google Android accepts RISC-V,It’s like RISC-V officially entered the Andriod ecological arena,Before, it was more like warming up on the sidelines. "Meng Jianyi said.
Actually, Andriod supports RISC-V, and flathead brother plays a key role.
In addition to the mobile ecosystem, RISC-V is also moving towards the PC market with higher performance. At the Yunqi Conference in 2022,Pingtou also announced the completion of the first adaptation of RISC-V to the community version of Tongxin operating system., successfully running desktop environment and a variety of office and development software.
Not only that, not long ago, based on the high-performance SoC prototype "Drag Shadow 1520", Pingtou completed the adaptation of the dragon lizard operating system, which is the world’s first RISC-V architecture compatible with the operating system on the cloud.
From the terminal to the PC and then to the cloud, Pingtou Brother has completed the deep adaptation of RISC-V with RTOS, Yocto Linux, Android, Dragon lizard, Tongxin deepin and other international mainstream and domestic operating systems.

It also means that,Brother Pingtou has removed the obstacle of RISC-V developers developing end-to-end cloud software system adaptation,Although there is still a lot of work to be done, at least the first pioneering step has been taken, which provides a good foundation for the next RISC-V ecological prosperity and rich application.
"The ecology of RISC-V has established a certain foundation, both in general computing and in computing with limited resources." He Xiaoqing, director of China Software Industry Association and chairman of Embedded System Branch, thinks.
Meng Jianyi believes that when 100 billion RISC-V chips are shipped, the ecology of RISC-V should have been basically established by then, which can also prove the tenacious vitality of RISC-V architecture.
With the gradual completion of hardware and the accelerated development of software ecology, the key to the surprising application of RISC-V lies in grasping the development direction of RISC-V.
03
RISC-V should also be in 5G mode.

Ni Guangnan, academician of China Academy of Engineering, said at Yunqi Conference 2022: "The open source model has moved from the software field to the hardware field, and RISC-V provides an opportunity for China to take the initiative in the development of chip industry. China should play its due role in the global open source field, and contribute China’s wisdom, China’s program and China’s strength to break the technological monopoly and promote the world’s collaborative innovation. "

Obviously, the core of RISC-V is open source,To stabilize the RISC-V architecture, different companies need to explore it from different angles, otherwise the architecture will have no vitality.At this time, contributing more institutions or companies will play a more important role in the future development of RISC-V.

This is also an opportunity for China Company. Bao Yungang, a researcher at the Institute of Computing Technology, Chinese Academy of Sciences, believes that,In the past 20 years, there are two main modes of CPU development in China.One is through "introduction-digestion-absorption-re-innovation"High-speed rail mode;One is to build a technical system independently and build an industrial ecology independently.Beidou mode.

In recent years, the third mode-5G mode has appeared.More and more domestic enterprises spontaneously choose open instruction set RISC-V, participate in the global ecological construction of RISC-V, and build international advantages by integrating into the international ecology, being compatible with international standards. "Bao Yungang said.

Under the 5G mode of CPU, it is even more necessary for China company to actively participate in the formulation of RISC-V standard and contribute more to RISC-V community.

"Limited by the concept, company size, talents and other conditions, domestic RISC-V companies have relatively less investment and contribution to the open source community than many foreign companies." Many people in RISC-V industry said to Leifeng.

"Our software team has slowly stepped out of a path and will make a certain contribution to the software. We will gradually have such a system, which has a great investment.We also call on domestic companies to increase their investment in RISC-V field."Meng Jianyi said.

It is reported that Pingtou is a member of the board of directors of RISC-V International Foundation, and leads 11 technical directions in the foundation, such as data center, storage management, Android, security, etc. Its importance and quantity rank first in the world and the first in China.

Just as both x86 and Arm have encountered accidents in the course of development, no one can know what kind of surprise RISC-V can bring to the world at present, but only by making greater contributions to standard setting and ecological development can we have a better chance of making history.

Will Alibaba’s forming rich ecology of cloud integration bring some revolutionary changes?

//

Artificial Intelligence Governance and Practice of Sustainable Development

This white paper comprehensively summarizes Alibaba’s practice in the field of AI governance and sustainable development, focusing on the current hot issues in AI application, and systematically introduces our practice ideas and methods from the aspects of data, technology, management and multi-collaboration. At the same time, it is explained with some special topics, hoping to provide useful reference for all sectors of society.

At present, artificial intelligence technology is developing vigorously, widely empowering thousands of industries, and bringing profound changes to human production and life. While artificial intelligence promotes social development, there are also risks and challenges.

A large number of artificial intelligence governance and development principles have been formed all over the world. The next step is to put the abstract principles into practice. As an important force in the research and development and application of enterprise artificial intelligence technology, it is urgent to explore a set of artificial intelligence governance practice system suitable for its own business development, implement all governance requirements into the whole life cycle of artificial intelligence, and lay the foundation for the release of artificial intelligence dividends with effective governance, thus accelerating the realization of sustainable development vision.

AI001: White Paper on Artificial Intelligence Core Technology Industry

AI 002: 2021 Love to analyze the application trend report of artificial intelligence

AI003: Data Security Risks and Governance of Artificial Intelligence

AI 004: Ten Technological Advances of Artificial Intelligence in 2020

AI 005: 2020 Panoramic Report of Artificial Intelligence Manufacturers in China

AI 006: 2020 China Artificial Intelligence Business Landing

AI007: White Paper on the New Generation of Artificial Intelligence

AI008: The Way to Win in the Future of Artificial Intelligence

AI009: White Paper on New Infrastructure Development of Artificial Intelligence

AI010: Tencent Artificial Intelligence White Paper

Typical case of ai 011: 5 Gai

AI 012: AI+White Paper on the Development and Application of Higher Education

AI013: China AI+ Retail Industry Development Research Report

AI014: China AI+ Retail Industry Development Research Report

AI 015: Panorama of AI Industry

AI 016: 2020 China AI Application Trend Report

AI 017: 2021 AI Intelligent Manufacturing Research Report

018: 2021 China AI Commercial Landing Market Research Report

AI 19: 2021 Research Report on the Development of Cognitive Intelligence

AI020: Shanghai Artificial Intelligence Innovation and Development Exploration and Practice Case Set

AI 021: 2021 AI China-Taiwan White Paper

AI 022: 2021 Global Research Report on the Application of Artificial Intelligence Education

AI 023: 2021 White Paper on Trusted Artificial Intelligence

AI024: China AI+ Security Industry Development Research Report

AI 25: 2021 White Paper on Intelligence in the Cloud

AI 26: 2021 Research Report on Artificial Intelligence+Medical and Life Sciences Industry in China

AI027: China Cloud Native AI Development Platform White Paper

AI028:2021 Application Research of AIoT in China Smart City

AI029: White Paper on Empowering Urban Space Management in China AI Middle and Taiwan

AI030: White Paper on Cognitive Neural Basis of Artificial Intelligence

AI031: Path, Method and Leading Practice of Enterprise Intelligence

AI 032: White Paper on the Development of AI Framework (2022)

AI033: White Paper on Artificial Intelligence (2022)

AI 034: 2022 Top Ten Trends Report of Deep Synthesis

AI035: China AI Digital Commerce Industry Outlook 2021-2025

AI036: Explainable AI Development Report 2022

AI 037: 2022 White Paper on Baidu Artificial Intelligence Patent

038: Development status, application scenarios and typical enterprises of AI+digital twins

AI 039: 2022 Global Artificial Intelligence Industry Research Report

AI040: Digital Innovation of Neuroscience in China (2022)

AI041: White Paper on the Supporting Environment of Artificial Intelligence Education in Global Primary and Secondary Schools

AI 042: 2022 China Knowledge Map Industry Research Report

AI043: Panorama of Knowledge Points of Artificial Intelligence

AI044: Artificial Intelligence Generated Content (AIGC) White Paper

AI 045: 2022 China AI Commercial Landing Research Report

AI046: China AI Technology Application Scenarios Market Research and Selection Evaluation

AI 047: 2022 Financial AI Development Research Report

AI048: Report on the Ecological Development of Trusted Artificial Intelligence Industry (2022)

AI 049: 2022 China Dialogue AI Industry Development White Paper
AI050: White Paper on Application Practice of Artificial Intelligence Technology
AI051: White Paper on Artificial Intelligence Standardization (2021 Edition)
AI052: Annual Report on Ethical Governance of Artificial Intelligence in Shang Tang (2022)
AI053: An Efficient Investigation Report of Artificial Intelligence Deep Learning Course
AI 054: 2022 White Paper on the Development of Artificial Intelligence

AI 055: 2022 White Paper on Medical AI Industry Research

AI056: White Paper on Practice of Artificial Intelligence Governance and Sustainable Development

The following is part of the report.