Chairman, Huawei
Huawei: Live From Huawei Connect 2019 Day 2 Keynote
We're streaming live from Shanghai Expo Center. Don't miss the highlights of Day 2 keynote at Huawei Connect 2019.
About Liang Hua
On May 25, 2026, Liang Hua delivered the opening speech at the East Coast 2026 event, where Huawei presented the Tau (τ) Scaling Law. He stated that over the past six years, his team had "explored new paths in semiconductor and find a way for sustainable evolution." Liang Hua described a shift "from geometric to the time scaling as the new guiding principle for electronic system evolution." He said that by 2031, Huawei's high-end chips based on this law are expected to feature a transistor density equivalent to 14 Å (1.4 nm) processes. Liang Hua discussed the company's progress, noting that "logical folding is a new but universal design methodology for digital circuits and systems" and that "over 80% of energy goes into moving data and over 70% of cost into storing it." He said that the Tau scaling path has been proven "feasible universal and sustainable at the different time levels" through six years of practice involving 381 chips. Liang Hua added that the company will "bring the surprise not saturation not continuation but a big leap ahead" and that its vision remains "to bring digital to every person, home and organization for a fully connected intelligent world."
Transcript (12 segments)
✨ AI-enhanced transcript with speaker attribution
L
Liang Hua0:25
These Lemax ladies and gentlemen, friends. Good morning, welcome to HUAWEI CONNECT 2019. Yesterday, Mr. Kan who articulated Huawei's computing strategy. Today I will continue with a talk; I will share with you our thoughts in computing industry and the plans we have. The spectacle behind me is the top of D5, one of the five famous mountains in China, Tai Shan. You can see the rising sun and a clamp on spreading its wings. It is our expectation that Huawei's computing industry can be like a clamp on spreading wings, flying up in the sky. On so many levels, computing has accelerated humankind's knowledge of the world. For example, to complete the genome sequencing, it took 12 years in 2003, but now it only takes one day. The reason behind it is because the computing power is 1,000 times faster. 20 years ago we were only able to forecast the weather one day in advance, but now we can have a pretty good accurate weather forecast for seven days. It's because the computing power has improved twenty thousand times. Another example is that humans with naked eye can observe the galaxy 2.2 million light-years away from us, but now China's FAST telescope is able to observe the outer space as far as 13.7 billion light-years away. Currently, the entire computing industry is undergoing profound changes, and we believe there are four trends. The first trend is the change from data center to computing center. Now, in the hyper-scale data centers, the cost of servers continue to go up. For example, if you look at our cloud at DC after our optimization, you can see that the costs incurred by servers now make up 60% of the total cost of the DC. It means that computing is now becoming the major pillar of the DC. At the same time, we see annual growth of AI computing demand is growing 10 times every year. It is expected that by 2025, AI computing will make up 80% of the DC computing power. In order to fully tap the potential of data, I think that we have to switch gear. In the past it was all about storing the data, but in the future we think that computing plays a more important role. That is why we think that the trend is the data centers are changing to computer centers. The second trend is device and edge driving innovations in the computing architecture. In 2018, 23 billion on-trip chips were shipped around the world, mostly for device and edge. Also, in the same year, general-purpose CPUs were shipped less than 30 million units. It's a difference of hundreds of times. It is forecasted that in 2019, the total computing power of smartphones shipped by Huawei will be twice as big as the additional computing power in global DCs of the same year. It is an indication that edge and device computing power will drive the development of the entire computing industry, which in a way will drive innovation in the central computing architecture. At the same time, because of the high bandwidth and low latency of 5G, more applications on mobile will be on the cloud. For example, mobile games, VR adventuring will be on the cloud, and interaction will happen on the device. At present, there are over 5 million applications on smartphones. However, data centers primarily based on x86 servers, in order to migrate the smartphone applications to the cloud, and if the architecture is different, you will have a loss of performance of 40%. This is not just a hindrance to the user experience, it's going to add to your cost of migration. However, if your device and cloud architecture is homogeneous, you can address the problem from the very beginning. The third trend is greener computing. Yesterday we launched the Atlas 900 cluster, whose computing power reaches 256 petaflops. According to our calculation, if you use general purpose CPU for the same task, it will need 6,195 cabinets; if you use a GPU for the task, it will need 208 cabinets; if you use an NPU for example, Ascend, you will need 128 cabinets. This is all attributed to the optimization because of the use of deep learning technology in Ascend architecture. But ultimately we managed to use 16 cabinets only. It's not just the chip level optimization; we also have system level optimization. For example, the board grade liquid cooling and enclosed heat dissipation, heat isolation. Power consumption is reduced from 40,000 kilowatts to 736 kilowatts. We think that computing is going to be greener and greener, which is crucial. The fourth trend is computing architecture will continue to see more innovation. In the past five years, general purpose CPU has met a lot of bottlenecks because performance per core has improved less than 10%. You can see that is approaching the ceiling of Moore's Law. Now we are running a race and we're almost at the ceiling. Now we have to break the different walls: efficiency wall, Moore's optimization, memory wall, and high-speed I/O wall, in order to get to the next level, so that we can tap the potential of computing. That is why we believe the computing industry has entered into the golden era of architectural innovation. In order to meet the computing demand which is growing rapidly, computing industry needs to move forward. Abundant, high performance, diversified, and greener computing power would be the major drivers for the development of an intelligent world. Huawei always stays customer centric, so the customers' requirements drive our way to continue to innovate in computing architecture and engineering technologies. Our pursuit is to provide computing power for ubiquitous cloud and pervasive intelligence. In order to realize our pursuit, we are dedicated to building one cloud and two engines. These are the major pillars behind the open ecosystem, and it is best illustrated by the plane in the picture. We've got two engines which are Kunpeng and Ascend, and we have chipsets, and this supports heterogeneous computing architecture. Intelligent computing and intelligent data storage are the two wings. Cloud, edge, device coordination is going to deliver diversified computing power with a combination of Kunpeng plus Ascend plus x86 plus GPU. In terms of intelligent data storage, we will have integrated solutions of storage, Big Data, database, and artificial intelligence, looking at the whole life cycle of data. We would like to make sure why the cost per bit is minimal; we would like to attain the maximum potential of data. The one cloud refers to Huawei cloud services. We are dedicated to full stack innovation, and we provide hybrid cloud services. We become the black soil for the ecosystem partners by providing affordable computing power. Open ecosystem refers to our approach to open hardware and open source software that enables partners. We want this to be a healthy and robust industry ecosystem. So this is our overall one cloud, two wings, two engines, plus ecosystem strategy. As you know, processors are the foundation of the entire computing industry. You will only have computing power when you have the processors. Huawei started its investments in embedded processors in 2004; it's been 15 years. After now, we have already established our fundamental chipset families centered on Kunpeng and Ascend. Kunpeng includes server and PC chips. Kunpeng 920 is the industry's first 64 core data center processor. Its performance is 25% higher than the industry's mainstream processor. Its memory bandwidth is 60% higher than the industry mainstream counterparts. At the same time, we've got a 4-in-1 SoC with CPU, bridge chip, NIC, and SSD. It is the most integrated data center processor in the world. Ascend has included training and inference chipsets. Ascend 910 is used for training. You know, the Atlas 900 cluster has used Ascend 910. It's a half precision, has reached FP16, which is twice as high as the industry. Ascend 310 used for inference is an integer precision, has reached 16 Tops, with a power consumption of 1 watt, while the industry averages 10 watts. In addition to that, our SSD chips have a unique wear leveling algorithm; it has extended the life of our SSD drive by 20%. This is far leading the industry. Our intelligent NIC chips can support offloading of the processing of various protocols such as OVS, VXLAN, and RoCE, especially suitable for cloud and low latency scenarios. We also have intelligent management chips which can support intelligent fault management and power management. This is the foundation of automation in hyper-scale data centers. In the current computing architecture, Huawei is the only one that has all the key chipsets, including CPU, NPU, intelligent storage controller, and network interconnect, as well as intelligent management. We hope that through innovation of the whole portfolio, we are able to drive the computing sector forward, so we can enjoy another golden era of the computing sector. For chipsets, where we have one generation in mass production, we are researching and developing the next generation, and planning for the next next generation. Since 2007, when we first invested in Kunpeng, it has been 12 years. Kunpeng 920 is the third generation chipset. Kunpeng and Ascend series will continue to evolve in the long term. We plan to launch one generation every year to continuously enhance competitiveness and also ensure backward compatibility. We hope that through this arrangement, we are able to drive the computing sector forward. This slide is about how with positioning of the computing sector, our strategy is to open hardware, open source software, and enable partners, and promote to the computing industry. Huawei will focus on innovating in computing architecture, researching and developing processors, and also providing cloud services. Using Huawei's hardware capabilities, we will be able to offer boards, SSDs, NICs, RAID cards, Atlas modules, and cards to support partners to develop servers and PCs of their own brands. Our Taishan servers target the high-end market. We also provide some components for Huawei products when conditions are right. That means when the partners are ready, Huawei may stop selling Taishan servers as a standalone business. What we want to make is database and AI computing framework available through open source so as to enable partners to develop their own products. We also want to provide a development framework for developers that covers all the scenarios including edge, device, and the cloud. Huawei does not develop applications ourselves, but we support applications and software to be migrated to our platform. Here is the ecosystem of the IT industry. We estimated that by 2023, in the whole world, this sector will have a market space of around 2 trillion US dollars, and in China, it will be over 1.1 trillion RMB. Kunpeng and Ascend, like the roots of a tree; only when the roots are deep, the leaves will grow. Huawei will continue to invest in Kunpeng and Ascend R&D. We also hope to work with partners in the whole value chain to grow the IT industry and share the dividends of the digital world. I want to stop for a while and give the floor to my colleague, Mr. Mahesh, who always has insights on the computing industry and Huawei strategy. Please welcome President of Huawei Intelligent Computing Business Development, Mr. Michael Ma, for his presentation: Innovating for the Ultimate Computing Power.
M
Michael Ma18:44
Ladies and gentlemen, good morning. This morning we witness some very important announcements. I'm very glad to be on behalf of Huawei's computing department to announce a series of computing products. The computing sector is embracing another golden era. In the general-purpose computing domain and AI computing domain, we have made strategic investments continuously to provide the ultimate computing power to the world. We know that steam power drove the first Industrial Revolution, achieving the transition from agricultural civilization to industrial civilization. Electricity powered the Second Industrial Revolution, which significantly enhanced production efficiency, and human society entered into the electrification era. And the third Industrial Revolution represented by computers and information technology, as well as the one represented by AI, computing power is becoming more and more important. In the future we enter into an intelligent world, AI will be pervasive. Autonomous driving will be adopted by tens of thousands of households. Smart robots will fundamentally change the way people live and work. Behind all of these intelligent applications, we need powerful computing. Computing power is the pivotal driver of the intelligent world. Our aspiration is to provide the ultimate computing power and solve world-class problems, and become a key force to drive the progress of the intelligent world. The intelligent world needs products with ultimate computing power. So what does such a product look like? In the general-purpose computing domain, we believe such a product should have three characteristics. First, multi-core, high concurrency. Therefore, for a single processor, 64 cores is a basic configuration. Second, real-time big data analytics and distributed databases need to exchange massive data with memories. Therefore, 8-channel memory is a must. Third, the collaboration between CPU and an accelerator requires high bandwidth and low latency I/O. It is imperative to upgrade the bus capability to PCIe 4.0. But this is not sufficient to meet all the requirements. The current computing architecture is evolving from a centralized one to a distributed one. If we only have a powerful CPU, it is not sufficient. We believe this server needs to have a 4-in-1 SoC, XPU, high-speed interconnection so that we can have powerful computing from CPU to service, and achieve ultimate computing power flow of clusters with 100G high-speed IO. Next, I'd like to explain more details. We often say that whether a car can drive fast or not depends on the engine. For a product with ultimate computing power, it must have a powerful processor. Kunpeng processor integrated 64 physical cores. According to SPEC assessment, it got 930 points, 25% higher in terms of performance compared with leading products in the industry. Kunpeng processor also adopted a 4-in-1 chip architecture. It's not just one CPU. It has integrated RoCE and SoC controller chips as a bridge to chipsets. A single processor has integrated full chipsets, so it is more dense and also has lower power consumption. We developed our own cache coherency bus, HCCS. It can achieve high-speed CPU connection. The communication data rate can reach 30 GT/s. Also, through CPU interconnection, we have achieved 256 physical cores NUMA architecture, and we are able to provide a full serve server which is ARM compatible. CPU and NPU interconnection is also very critical. Huawei has applied HCCS for interconnection between CPU and NPU, therefore we use cache consistency. But for XPU, XPUs talk to each other directly. Based upon this architecture, we can have a flexible combination of general-purpose computing and AI computing. Therefore, we are able to provide a heterogeneous computing server with the most powerful computing capability. Current processors usually use external NIC to provide 10GE and 25GE interfaces. For service under a distributed architecture, we need, in order to have a high-performing cluster, we need to have high IO throughput. Kunpeng server is the first one in the industry to have a general-purpose processor embedded with 100G network mobility. So from processor to server, to the cabinet, and computing cluster, we've achieved 100G high-speed network connection. Servers have entered into the 100G era. Just now I defined products with ultimate computing power: they need to have a 4-in-1 SoC, XPU, high-speed interconnection, and 100G high-speed IO. So some of you may be wondering, is there such a product that meets all these requirements? The answer is Huawei's Taishan server. With Taishan server series, we cater to big data, distributed storage, database, HPC, ARM native applications, etc. It is compatible with some architecture. Huawei offers storage-intensive, compute-intensive, edge computing, and some other models of server products. So data center and edge as well as other scenarios can be addressed. Today I'd like to announce a very, very important product of the computing sector. Please enjoy the video. I want to announce the official launch of Kunpeng Board, and it is accessible to our partners. Kunpeng Board carries the dream of many colleagues in the computing sector. This board carries two Kunpeng processors. It has 128 physical cores. It is embedded with 100G RoCE and NIC. It has 32 memory slots. It also supports PCIe 4.0. Partners can use this Kunpeng Board to develop computing products of different formats. Thank you. Thank you, my colleague Polly has over 30 years R&D and manufacturing experience regarding hardware. We have shipped over 1 billion pieces of board. Kunpeng Board has 56G high-speed service capability. The performance is up by 25%, and BER is less than 10 to the power minus 12. The thermal design power is 15% lower than industry average. By using dynamic energy management technology, the energy efficiency ratio is also 15% higher than the industry average. Partners can develop products based upon Kunpeng Board and you can enjoy high performance, high reliability, and high efficiency ratio to fully tap the potential of the equipment. If we work together, we can achieve more. Huawei will support our partners to develop more products with ultimate computing power and offer more and better choices for the customers. Kunpeng Board has embedded BMC chip and BIOS software. We will provide board interface spec and equipment management aspects to enhance the equipment efficiency and quality. Huawei will also make our hardware engineering capability accessible to our partners and provide design guidelines to our partners. In addition, we also provide a compatible list of memory hardware, NIC, OS, etc. Our partners can refer to the design guidelines and use Kunpeng Board to quickly develop your own server and desktop products. And now I want to invite our partner to give a presentation. Please welcome the CEO of Tongfang company, Mr. Lee Jin-hyung. He'll share with us about how to create high quality computing products.
L
Lee Jin-hyung31:02
Ladies and gentlemen, good morning. It's a great honor to be at HUAWEI CONNECT 2019. I'm from Surat Anfal. My name is Lee Jin-hyung. I'm in charge of the operation of the company, and at the same time I'm the general manager of Tongfang PC business. Tongfang was initiated by my university, a high-tech company established in 1997. It's been 22 years of business. Our major business is IT environmental protection. The PC business has always been one of the mainstream businesses from the company since the very beginning. We have made strategic deployments for our PC business. Now we are embracing the new world. Kunpeng processors and Kunpeng server boards have just been launched by Mr. Mahesh. We have new technologies to build into our products and solutions. We think that we can all look forward to it. We've got 22 years of experience in the computer business, and we have laid down a very solid foundation, especially in our in-house R&D. We've got really strong R&D competence, while at the same time we've got strategic presence including manufacturing, services, after sales, and marketing. We have strong capabilities across the value chain. Now let's have a look at some of our numbers. Tongfang computer is now the second largest integrated equipment provider in China. We're also providing servers, but that's not our major business, but anyway, we are the second largest integrated computer equipment provider in China. We've got over 1,000 R&D engineers that can guarantee in-house development for the whole lineup of PC products. We have 30 different categories of computer products in our portfolio; it can cover different segments from individual consumers to enterprise customers such as government, education, healthcare, and financial institutions. You name it. We have production bases across China in Jiangsu, Shandong, and Guangdong, and northern parts of China. In total we have over 10 million units of manufacturing capacity across China. At the same time, we're proud of our service teams. We've got over 3,000 service personnel across the country, and that has established our strong position in the China market. Huawei is a global leading tech company as we know. Huawei has been dedicated to the communications industry, but in recent years, Huawei has given great attention to the IT tech development. And Tongfang has followed up. Since the beginning of this year, Tongfang has engaged with Huawei, and we have assessed the Kunpeng processors and boards and the prototypes, and together we have tested the solutions and researched and developed the integrated solutions together. From the test results, we think that it's pretty satisfying. The design of the board and the layout is pretty reliable and stable because Huawei offers a server board solution, so it's got good compatibility with software including BIOS. From this angle, its components are very mature because we can readily integrate it in our solutions. The server boards are a pretty good base; it's very easy for us to integrate it in our solution. We can efficiently deliver the ultimate integrated equipment to our end users. As I was saying, we've got 22 years of experience in the computers industry; we are very strong in the design of integrated computers. To work with Huawei's Kunpeng boards, Tongfang has come up with a special design for high-end premium customers. It is a special cabinet, specifically for the high-end business customers. It is easy to dismantle and easy to install. We've got a lot of reliability designs like vibration-proof and radiation-proof designs. Mechanically we've got strong competence, and electronically we draw from the strengths of Huawei's Kunpeng server boards. Because of this, we are able to promote Tongfang's computer based on Kunpeng server boards. Here are the defining features of the desktop computers. First, high performance enabled by Kunpeng server boards. These are the results of our optimization with Huawei. You can have a look at this: for a single-core, we have run a UnixBench test; single-core score reaches 1491 points, multi-core reaches 6049 points. Such performance has caught up with the mainstream CPU performance in the world. In addition to that, it is ultra reliable because we have added to it a TCM crypto technology chip, so it is able to guarantee information security. It is also equipped with reliability design as I talked about: vibration-proof, radiation-proof. It is able to deliver a seamless experience. The desktop computer based on Huawei's Kunpeng server boards is able to play 4K video very smoothly. Tongfang has established great partnerships with Huawei. But it's just the beginning. We have just launched Tongfang's desktop computer based on Kunpeng server board. We look forward to more in-depth cooperations between the two companies. I look forward to more technological collaboration. Right now we are carrying out system-level collaboration, but in the future I look forward to going deeper into the server board level or even deeper in the system, to include customization where we can contribute value. We have great confidence and capacity in our company, so we are dedicated to defining more categories of products, not just limited to desktop computers, but also we have great potential in developing PCs, tablets, and other form factors of consumer devices. We look forward to closer cooperation with Huawei. At the same time, we stay close to the customers' requirements. Tongfang computers based on Kunpeng will deliver better user experience. And I think that we are reinforced to work with each other in developing the right sector for applications and software. For example, we can embed AI capabilities to have retina, fingerprint, and brain recognition. So we see great untapped potential of Kunpeng computers, so that will shorten the gap with global mainstream computers. We are dedicated to delivering better computers and services to the consumers. We would like to work with Huawei and contribute our fair share to the Kunpeng ecosystem and also to the IT industry development of the country at large. Last but not least, I wish the Kunpeng system better and better, and also wish all Huawei a greater future.
M
Michael Ma41:22
Thank you, Mr. Lee Jin-hyung, for your wonderful sharing. In addition to general-purpose computing, Huawei has made intensive investments in innovative AI computing. So what are the key defining features of the most powerful AI computing? Different from inference, the chips for training are the best testament of AI computing power. Ascend 910 is based on Ascend Da Vinci architecture. It is equipped with the industry's 3D Cube computing engine. One single engine is able to complete 4096 operations of multiplication and addition, and its computing power reaches 256 teraflops. It has millisecond craziness and on-device processing. It can support multi-chip training. Our servers have been equipped with eight Ascend 910, reaching 2 petaflops of computing power. Through scaling out, we can expand it into AI clusters. What we launched yesterday, Atlas 900, is a scalable AI cluster. It is basically a similar blockless network technology integrating chip, servers, and clusters. It can reach 1024 petaflops. The training time can be shortened from months or weeks to days and even seconds. On August 23rd, Huawei launched the industry's most powerful AI training processor called Ascend 910. And today I would like to announce the latest products for AI training: they are the powerful Atlas 300 training card and also Atlas 800, the training server. Atlas is the name of the god of strength in Greek mythology. We gave the name Atlas for our AI computing products. It is our expectation that Atlas will become the solutions that provide the most powerful computing capabilities. Next, I will talk more about our two new products. This is what Atlas 300 looks like. The industry's most powerful AI training card, Atlas 300, provides 256 teraflops of computing power. It is twice as big as the industry mainstream card. It can train 1802 images per second, better than the 965 images per second. It supports 100G RoCE high-speed interface. It can realize parallel transmission of gradient parameters and data transmission. It can reduce the transmission time by 70%, so that training time can be shortened to second level. Atlas 800 is the industry's most powerful AI training server. It is equipped with four Ascend 910 AI processors, providing 2 petaflops of computing power. Its computing density is 2.5 times higher than industry average. It is only 75 kilograms, half as heavy as the industry counterparts. It has 32 hardware decoders. Every second it is able to decode 16,384 images, and these images are 1080p, and that means it is 25 times stronger than the mainstream products in the market. In addition to that, it's got parallel processing capability. It supports two heat dissipation methods: wind and liquid heat dissipation, that satisfies diverse needs of enterprises, DCs, and clusters' high density deployment. Its power efficiency is 1.8 times better than industry. We fly our own parachutes. That is why we have deployed Atlas 800 in our DCs in Sunshine Lake of Dongguan. We have fully deployed a fully liquid cooling system, and it can achieve heat dissipation as much as 550,000 watts. Our Atlas 800 has eight Atlas 800 surfaces in one cabinet, and that is why it is able to support a capacity of 50,000 watts per single cabinet. Yesterday we also launched Atlas 900, the cluster, and this went viral in our moments and also in social media. Atlas 900 is a scalable AI cluster architecture composed of Ascend 910 processors. Thousands of Ascend 910 processors in a ResNet-50 image test scored 59.8 seconds, number one in the world. With the same precision, it is 15% faster than the second runner. Atlas 900 is equipped with super powerful computing power which can be used for scientific research and business innovation. For example, the cases we shared, Atlas 900 has been used for astronomical exploration and oil exploration. I believe those are very impressive cases. Atlas 900 can be used in many areas in the future. Ascend 910 and Ascend 310 based AI processors are part of our whole portfolio of Atlas products. It means that we are able to provide strong computing power for training and inference. It supports all scenarios. It is based on Da Vinci architecture, and it supports cloud, device, and cooperation that can speed up digital transformation of different sectors. This year we are collaborating with the Southern branch in innovating the inspection of the equipment. From the screen you can see our cooperation. Look at the comparison before and after we use Atlas technology in the inspection. What we were told by the customer is that one engineer needs to travel a long distance, and he has traveled through the most cold mountainous areas, and it is sometimes very dangerous for him to do the inspection of the grid. But now based on Atlas, it is fully automated inspection, so that the customer no longer has to do it manually. They don't have to go through the dangerous mountainous areas anymore. They can have an automated system that can support real-time alerts and accurate reporting. Technology is for good. Pass it on. We hope that more and more industries are able to leverage technologies to make the world a better one. Next, I'm handing over to Mr. Hussein, the CTO of ABB. He is going to share with us the industrial case of the Atlas solution.
B
Bosaley Hussein50:13
Good morning ladies and gentlemen. It's really exciting to be here. Yesterday, pronounced in Hall H, it was really satisfying to see the interest at the sight of an industrial robot separating garbage. China Daily carried it in a news story, CCTV had a video clip on it. So it clearly shows that the interest in both the topic and the solution is really wide. Today I'm going to be talking about some of the challenges that our industry are facing, introduce industrial autonomy and its relation with AI, and round up with a few examples. ABB, as my colleague EDO who is the Chief Digital Officer for ABB mentioned yesterday, is a 130-year-old company which makes a lot of things for primarily two industries. We serve the energy industry, the electrical energy industry, and the industry for process, discrete, and hybrid. We have done that for 130 years, and we've been a technology leader in both these segments. In the energy industry, for example, ABB is the company that invented and introduced high voltage DC transmission. In the automation segment, industrial robots were also an innovation that was brought into the market by ABB, and that was way back in 1974. So we have a history of being innovative, also being leaders in that, and being ahead of the time. Today, when we see the not really, we see that in both the industry, there are major shifts going on. On the energy side, a revolution is changing the way how electricity is generated, how it is transported, and how it is going to be consumed. On the industry side, we are going through the fourth Industrial Revolution now. One thing that is clear to us is that for these revolutions to succeed and deliver the value that is expected of them, we have to go beyond automation. Why do we need to do that? Because automation fundamentally has to be told how to react to every situation it encounters. Given the growing complexity that is developing in the industry, we need systems that not only react as they are programmed, but are able to handle situations that are not planned. So systems that have the capability to deal with unplanned situations in a proper manner, these systems are what we call autonomous systems. We do believe that the cornerstone technology for moving towards autonomous systems is going to be artificial intelligence. Artificial intelligence in the last less than a decade has made tremendous progress. It has shown superhuman capability in many areas such as games, whether it is a game of Go or Starcraft. I think it has clearly beaten the best of human players in that. However, and it is also for industry, there have been a lot of areas where we have been using that quite successfully. When it comes to condition monitoring, when it comes to predicting expected lifetime or time to failures, artificial intelligence has delivered great value to industry. But for AI to really play a key role that it must for industrial autonomy, it has to learn how to deal with unplanned situations. It has to come from a closed world to reality. The picture on the screen shows that you can have several events that you definitely did not plan for or anticipate. So what does it take for AI to move towards industrial AI? I'd like to highlight some of the key elements that are needed for industrial AI. First and foremost, I think it's about the algorithms. Today's AI algorithms have an explainability issue: you cannot make out why the AI algorithm reached a particular decision. That combined with the brittleness of AI makes it very difficult to be used, especially in physical processes, and even more so when the consequence of a wrong action can be either very dangerous or expensive, or both. The second aspect is about the data. In industry, since about mid-80s, most of the systems, most of the process plants have what I call distributed control systems. ABB was one of the pioneers in that. So there is lots of data that is available, but industrial data is about a variety of different equipment. If you look at equipment such as a transformer or a robot, there is very little that they have in common. There are literally tens of thousands of different kinds of equipment in any process plant. So there is a lot of data, but great variety of data. The second aspect of our data is that you can have several years worth of data, however it is not unusual to find that you have data from a robot for about three or four years of operation and you see no failure. It is difficult to train AI about a failure when there is no data related to failure. So even if there is, they are very few. So these are equipment that are designed not to fail. So the unevenness of data. The third aspect about industrial data is that being in normal operation is not a differentiator in industry. Everybody expects that. It is really looking at an outlier if you have had a golden batch, to find out why you had the golden batch and make that the new normal. That is the challenge that you have to address in industry for the deployment. For the deployment part as well, it is deployed in very real-time applications, very constrained equipment. So even if you look at the inference times that are there in the chips, they are not adequate for real-time use today. So that also has to change. And then AI has to be deployed at all levels: at device level, at edge level, and at the cloud level. So deployment everywhere. Let me come back to say a little bit about this demo that you see of the solution in H2. This has essentially four elements. It's got a conveyor belt on which the garbage is coming. You have the vision system that detects the presence of garbage on the conveyor and sends that information to the intelligence unit. The intelligence unit classifies that object, whether it is metal or plastic etc., and then there is a robot system with a gripper that picks up the item and puts it in the right bin. This was done actually as a collaboration together with Huawei. I'd like to point out a few things. First of all, the entire solution that was developed from ABB side, the entire team is a China-based team. ABB has 17,000-strong people in China, and we have very strong R&D capability here. So this was a solution which was developed here together with Huawei. The entire solution was designed and developed in just a little over one month. So it is really done locally and done with speed. So it's an example of a great cooperation that we had with Huawei on this. Now let me say what is going to happen next. Let me take an example of what can happen in future, and I'm going to take the example of the mining industry. The mining industry is very crucial for economies because the mining industry supplies the raw material for progress, but mining is not always safe and productive and efficient as it could be. ABB supplies a lot to the industries for exactly improving their safety and the efficiency and the productivity of the mines. But today, most of the mines that are operated are operated in island mode, as you can see in the graphic on the left. You have where you are doing the extraction, where you are doing the mining. So these are things that are then fed into the processing side; these are all treated as islands. When we look at a mine of the future, one of the things that we see that will come about, and this will be a confluence of several technologies. The first thing you notice is to take humans out of harm's way, meaning take humans out of the mines where there is danger. So humans work in safe control rooms that are above ground. Through the use of digital twins, virtual reality, and augmented reality, they are able to see and control the process. It is all integrated. So if there is a problem in one segment of the plant, they know how to deal with the next segment in the operations. Humans are replaced by autonomous equipment. So both for moving the ore, for charging the mine for new blasting areas, this is all done autonomously. And then you have a problem, utilizing digital twin and diagnostic capabilities, you are able to actually identify and fix the fault much quicker and much better. For this kind of vision to happen, it takes a lot of technologies that have to come together. ABB believes that this is not something one company can do. This has to be done in partnerships. The message that I want to leave you with is that it is partnerships like Huawei and ABB that will build a bridge to the future of industrial autonomy. Thank you very much.
L
Liang Hua1:04:30
Thank you, Mr. Husein, for sharing your views on AI's bright future for industrial autonomy. I want to thank our two presenters for their sharing. Just now, my colleague Mr. Mahesh shared a lot of information and ideas. So I believe we should combine the soft part and the hard part in order to unleash full computing potential. As we know, database and operating system are the two most important links in the system. A processor is like the heart, and these two are the souls. Since 2010, Huawei has invested in the R&D of server operating system, EulerOS. Up to now, in the field of telecommunication, storage, and cloud, we have applied this EulerOS system widely, over 100,000 units. It has two defining innovative features. Number one, triple intelligence scheduling technology: kernel, cluster, and die-by which I mean triple intelligence gathering. It can allocate different processes on the most appropriate kernel so that the application can have the shortest latency and the broadest bandwidth. We can shorten the process concurrency latency by 60%. Secondly, smart auto-tuning technology. We know the normal system would run slower and slower as time goes by. So driver, kernel, virtualization, library which constitute a multi-level perception and learning, it can conduct automatic optimization combining over 7,000 system parameters. The system will run faster and faster as time goes by. The web performance has risen by 137% in our test, and it is constantly optimized. Now I announce Huawei server OS EulerOS to be open source. The name of this open source version is OpenEuler. Huawei supports its partners to announce their commercial version based on OpenEuler. We also support different industry sectors' applications and software to be migrated to OpenEuler OS. I'm also happy to announce to you that we are now cooperating with Deepin, NeoKylin, Tongxin, CIC, Joly, and other partners to prepare an open source community, which will be officially launched by December 31st this year. Today it's my great honor to have with me Mr. Jiang, who is the CEO of Deepin Technology, to share with us his ideas. Welcome Mr. Jiang.
J
Jiang Mei1:09:38
Good morning everyone. It's my great pleasure to be here to take this opportunity to share with you our efforts in open source OS as well as our cooperation with Huawei. A brief introduction of myself: my name is Jiang Mei from Deepin Technology. A brief introduction: the company is rooted in Wuhan, and we have open departments in Beijing, Shanghai, and Guangzhou. For open source, we have built the community for the server OS. We have had very in-depth development and research for the community version and professional version. We provide extensive service to our customers. For the Windows OS system, we developed a desktop graphical desktop environment and a large number of application software and system security features, providing users with an alternative to the Windows operating system. And in big data and cloud solutions under multiple processor architectures, Deepin OS is China's first Linux distribution version with international influence. Since 2008, we now have 33 languages with more than 100 mirror sites for Deepin by our supporters in all continents across the world except Antarctica. There are local Deepin communities in Italy, France, Russia, Brazil, and our enthusiasts can discuss and develop in our community. Our digital environment and our applications, we call them in the family album of Deepin. They are also ported to more than 10 international Linux communities like Fedora, Ubuntu, and Arch. I remember when I was young, I was told that Chinese people should go beyond our national border and exert our influence across the region and in the world. I'm glad to see that Deepin has achieved that, and we always list top ten in the DistroWatch. We are rooted in China and we are serving the world. Deepin starts from open source and continues to give back to open source. Our technology community is the most active open-source operating system community in China, with nearly 100,000 registered users and nearly 1 million posts. Hundreds of patches were submitted to open source software such as Git, Qt, and Wine. It implements dozens of open source projects, provides more than 5 million open source lines of code, with more than 80 million operating system downloads. While we are engaged in the Deepin system based on the open source OS, we are thinking about the future direction and the targets of OS. From this chart, we can see that in different sectors, our influence and penetration rates are rising. From HPC, high-performance computing, we have already occupied 100% market share. We have the monopoly position in this sector. For the network server sector, we occupy 40% of the share. For the smartphone sector represented by Linux-based Android system, we have over 80% market share. Our only weakness is production terminal or desktop system. We have a relatively weak performance in this area with market share less than 5%. Looking ahead into the future, we believe that in the server sector we can maintain a stable position and continuously in the upward trend. For big data, AI, and cloud computing, the performance is becoming stabilized. But in the foreseeable future, we believe it is very likely that our market penetration rate of Linux OS will be over 50%. There are several issues including technical issues, but more importantly is about ecology. With China's information industry market development, and also with the advancements of technology, with the emergence of multiple device equipment, with introduction of interactive technology, we believe that China has the potential to be the dominant power in the next wave of markets. It will be a decisive factor in the coming 10 years. Looking back, we started from scratch, from a follower to an innovator, from x86 architecture to multi-platform. We have accumulated rich experience. From this chart we can see that in 2008 there was still learning by copying, but then in 2011 we were able to improve source code and develop patches in the mainstream desktop environment and also submit patches at a time. And then in 2014, based on WebKit, we developed the WebKit-based desktop environment. And then in 2017, when we found serious issues with the compatibility and performance, we rewrote our system based on Qt and GoLang in order to improve performance, compatibility, and user experience. You can see some screenshots of a few moments in the course of our development. In the product design and in the entire technology stack, we try to improve our usability, compatibility, security, performance, kernel, and we also cover the area of kernel, toolchain, driver, graphic library, and application service framework. This is our next version that we call V20; it's our next-generation version. We have made some revolutionary changes. For example, it can support PC and tablet at the same time, integrate different devices. It can also provide better security protection, including security system and device security. It will be the mainstream of our version. I believe this will be fully commercialized and offer consistent user experience, offering ultimate user experience in Linux terminal OS. We're very lucky that we are deeply rooted in the strong foundation of open-source community. We have achieved some tentative results. We feel also very privileged and happy to work with Huawei, such an excellent partner, to work together. We had very close cooperation with Huawei in terms of client OS for cloud platform and server OS for data center and computing center based on Kunpeng processor and Taishan server. We work together with OpenEuler system team to have the porting, adaptation, and optimization of the Deepin server operating system. According to performance benchmark tests, with the support of Huawei's Kunpeng processor and Huawei team, we can see that the performance indicators of many key applications have been significantly improved. For example, the big data software Hadoop and Spark have a significant drop in processing latency, while the throughput of the mainstream document database MongoDB has increased significantly. Different technology support opens the elastic community and component ecology. We cooperate with Huawei not only about technology but also about quality management, which is very useful for us. Starting from 2018, we kicked off our official cooperation with Huawei. We have experienced many important milestones together. In 2018 and 2019, we announced the Deepin OS version for Kunpeng server and established the commercial supporting service team. We can now fully support Kunpeng hardware and cloud computing environments. And also we have commercialized the virtualization and container technology. We work together with many domestic vendors to test and adapt in different scenarios, and actively organized and supported many POC projects oriented for finance and government sectors. We are delighted to see that Huawei's Kunpeng will unleash strong power in many unpredicted scenarios. The performance actually exceeded our expectation, as well as the expectation of our customers. This left a very deep impression on our customers. We also have finished the migration and certification on the Deepin desktop OS together with Huawei. We believe that the Deepin OS server and terminal products will soon be officially launched and widely utilized. I remember a person from Hebei once said, "The road is long and arduous, and I will never stop my pace." If we want to walk long, we need to work together. We have to work with many, including you, me, Huawei, Deepin, it includes everyone. We hope to become the leader of China's innovative ecosystem, and together we can embrace a brighter future. Thank you.
L
Liang Hua1:20:59
I want to thank Mr. Jiang for his wonderful sharing. As we know, the key applications of enterprises cannot be possible without a database. In the database field, Huawei has made investments over the past ten years. Now in telecommunication, finance, and other core scenarios, our database have been used widely. In May this year, we released GaussDB with two main features. Number one, it is the world's first AI-native, self-optimizing database. With the self-optimization capability of AI-native ability, it can improve the performance by 85% compared to DBA. Number two, it is also the global first enterprise-level database based on Kunpeng. It can fully utilize the hyper-parallel technology of Kunpeng, and compared with non-Kunpeng systems, the TPC-C performance has risen by 48%. So it is my great pleasure to announce to you that the standalone machine version of GaussDB OLTP to be open source. The name of this open source is openGauss. It covers 70% plus key scenarios of enterprises. We also support our partners to develop database products and applications of their own brand. openGauss will be officially launched before June 30, 2020. Now we open source our database and operating system, but that is not the end; it is only the starting point. We are going to go further and further, intensify our investment into these two fundamental software and others. We're going to do more R&D in these vast banks. We are going to partner with so many of you to co-develop these ecosystems to be even better and richer. Only by combining Kunpeng computing power, computing power hardware, and software, can we unleash the potential of the cloud and cloud services, and bring the computing power we need. Today I brought you here to launch Kunpeng-based cloud services. Watch the video. We have more than 69 Kunpeng-based cloud services and 43 Ascend-based cloud services now officially go live. These cloud services will be able to allow you to better experience the computing power brought by Kunpeng and Ascend. The multi-core and high parallel capacity of Kunpeng processors, and the interplay of software and hardware in cloud-native scenarios, the comprehensive performance can go up by 80%. When processing Big Data, the processing time can go down by 35%. Based on Ascend, the cloud server performance can be doubled or more, and be widely used for such scenarios as inference, training, autonomous driving. We hope that by full stack innovation, we can make AI inclusive. Today I have the delight to announce that for Ascend-based image search and invoice review, the price will be down by 70%. So these are the tangible benefits brought by technological progress. We also hope that we can make more of such announcements that amaze the industry. At the same time, Ascend is not only about price; it is also about value. For example, in the service sector, with knowledge graph, we have outperformed the amount of resource knowledge diffusion and boosted the efficiency of top knowledge experts in the oil and gas field to raise their productivity and efficiency by 70%. And for autonomous driving, our large cluster supports PB-level real-time data processing, helping the developers raise their efficiency by 5 times. To know more about the value brought by Huawei Kunpeng cloud services to the customers, please watch the video. You saw in the case study how Huawei cloud can be used to help the internet sector. Now I would like to invite the director of Beijing Municipal Bureau of Economy and IT to share his experience and practice in the government sector.
R
Representative from Beijing1:29:48
It gives me great pleasure to attend HUAWEI CONNECT for the convocation of this conference. I want to offer my warmest congratulations. Business leaders, thought leaders, and experts, there are so many of you as participants in this room. So I'd like to thank Huawei and the Shanghai city for giving us this opportunity to gather in the World Expo venue. I also thank all of you for your long-standing and persistent support and help to the informatization of my city, Beijing. From Summer Olympics to Winter Olympics, as the host city, Beijing has been developing itself into a digital city. What are we going to do for this? The Olympics motto is higher, faster, and stronger, and HUAWEI CONNECT also coincides with these aims of the Olympics. Since we want to host the Olympic Games, we have been serving the people to make life easier for the people. So we also want to make our services higher, faster, and stronger. From boosting the functionalities of the various authorities online and making the services online more accessible, we are going to make the city more intelligent offline and online. This is the biggest heritage of the Summer Olympics to Beijing. For example, before 2008, navigation was about from point A to point B, no congestion, because at that time people didn't know of congestion and noted the navigation software. But nowadays, everybody is a contributor and a victim of traffic jams. But now the navigation can provide the route. So in terms of intelligent services, roadside parking, ETC, and tolling can all be provided with strong support of computing power. You know, the Winter Olympics in 2022 can be 5G-based autonomous driving, playing the leading role. What will the Winter Olympics bring to Beijing and the country at large? Kunpeng is here to give you the answer. Thank you. As a pilot of deploying Kunpeng in Beijing, the first government cloud for Kunpeng took from decision-making to construction and to launch, it took us only three weeks, 21 days. It is about how we go faster from decision-making to application. There are seven application systems including the backend mobile services for common people and the accounting service, altogether seven application systems that were jointly launched. So we migrated from technologies to a homogeneous platform, so it took us only two hours to do the changes and migration. It tells you how easy it is to upgrade and migrate. So we have also deployed more than 2,000 Taishan servers, and we have also processed more than 600,000 DB, and the big data analysis efficiency is up by 30%, and the fault rate so far is zero. The interprovincial toll has been abolished among Beijing, Tianjin, and Hebei. We have used the Atlas servers to make sure that even if you drive on the highway at a speed of 220 kilometers per hour, the recognition accuracy will be more than 99%. But don't speed. Deploying Kunpeng in Beijing and setting up the joint Innovation Center have a good reason. First, because of the cultural heritage of Beijing. Beijing has been a capital city for 3,000 years, and it is inclusive, ethical, and international. It is committed to openness and innovation. Same for Huawei. Among Kunpeng, Huawei's systems got their names from Chinese classics and legends, and Gauss is a Western name, so people and things are called a name for a reason. So we have a nice name like a solid foundation. In Beijing, there is also a wealth of top experts and talents in cybersecurity, and the connections are all supported by the National Laboratories. Beijing has more National Laboratories than anywhere else in China. There are also leading universities and the talent bases. So Beijing has the best research resources, and we also have a complete industrial chain: chips by local companies, big screens by BOE, and the emergence of a big number of startups like Horizon Robotics, Face++, AI, and many others. Beijing has 35% of the unicorns in China. And also Beijing has one third of the top 100 Internet companies in China. According to the ranking of business environment in China, Beijing represents the top name in China. According to the World Bank ranking, Beijing is also high on the list. Therefore, Huawei would like to set up the joint Innovation Center so that the 16,000 IT firms in Beijing and all the use cases of the government can be fully empowered by this center. It is our hope that the center can be a hotbed for Kunpeng to thrive. At this place, I would like to also mention the Winter Olympics, a sub center of the capital city, and integration of Beijing, Tianjin, and Hebei. We are going to open all such possibilities into the center. Beijing's road is far away, and the far we choose is something. It is so timely for us to join here. So we are going to make a start from now. Imagine a vision: we stand in Beijing, look north, we are going to see the red leaves under Fragrant Hill. They are coming in the coming October. Looking west, we are going to see the merging point of five rivers and the bench in the tower monitoring this landscape. And that is where the sub center of Beijing is. There are so many headquarters of large companies and promising high-tech firms. When we pool our strengths, we are going to achieve good deeds. Thank you. I look forward to meeting you all in Beijing, and I would also like to wish HUAWEI CONNECT a complete success, share your insights into computing innovation and digital economy.
L
Liang Hua1:37:55
Thank you, Mr. Palin, he represented the height from Beijing. This journey will surely make a successful journey because it starts from Beijing. Just now we listened to the case sharing from our customers and partners of how they engaged with Kunpeng and Ascend. As always, Huawei is always the first user of the products we make. In our R&D internal IT products and solutions, our cloud, consumer cloud, we have used the products related to Kunpeng and Ascend. They have been deployed in more than 100,000 units. This year we will further improve the computing power of Huawei cloud and translate that computing power to clean computing power. Kunpeng and Ascend, we have many more cases. I would like you to watch the video so as to know about these cases. We need to make sure that Kunpeng and Ascend can provide better service. We need a healthy, robust industrial ecosystem, so we are going to make efforts in the following four areas. Number one, we are going to aggregate applications for industries. We will build a complete industrial ecosystem and competitive solutions with our partners in different sectors. Secondly, we will integrate industries in regions. We focus on processor and components development, but the computing industry is shared across the industry. We welcome our partners in different regions to develop your local versions of the Kunpeng computing industry. As a proverb says, a single flower does not make spring, while a hundred flowers in full blossom bring spring to the garden. We hope we can build a forest rather than a single tree. Number three, we will incubate standards with alliances. We will work together with Green Computing Consortium, AI Computing Consortium, and other alliances and organizations to formulate software and hardware standard systems for the healthy development of the entire industry and supply chain. Number four, we hope to empower developers through community building. We believe that developers represent the soul of the industry. In the next five years, we will work together with communities and universities to jointly cultivate 5 million developers, so that they can be the angel lifeblood to the industry of computing. To better support ecosystem development, in the next five years, Huawei will invest 1.5 billion US dollars to develop the ecosystem. The detailed plan will be announced tomorrow morning. So please refer to our agenda of tomorrow morning for more detailed information. We can see that the development of the computing industry is supported by the joint efforts of our partners in the industry. To better support the Kunpeng computing industry, we work together with ARM China, Green Computing Consortium, and other friends and partners to jointly launch the white paper on computing industry development. Now I would like to invite Mr. Allen Wu from ARM China and Madame Hu Guot from Green Computing Consortium to stand on the stage. Welcome. I'd like to invite you to share with us your thoughts on the computing industry. Mr. Wu, please.
A
Allen Wu1:46:20
Thank you for the invitation. It is my great honor to be invited here to attend the launching ceremony of the white paper on Kunpeng computing industry development. Every one of us may use three to four ARM-equipped devices, including your smartphone, the set-top box, your TV, and your automobile. So the open innovation system built by ARM has become the biggest computing platform and ecosystem in the world. In the next wave of science and technology revolution, you will see that 5G technology and computing power will enable the massive amount of data to transmit, and AR technology will translate data into productivity, bringing changes to every walk of life and our lifestyle. Together with Huawei, as early as ten years ago, we started our innovation journey. We applied ARM technology to the cloud. Ten years later, we are happy to see that Kunpeng is now in full blossom. In the next round of technological revolution, we hope to be rooted in local innovation and endeavor for global cooperation. The white paper illustrates our joint vision between ARM China and Huawei, as well as the technological structure and open ecosystem connecting cloud, edge, and device. This will become the biggest driving force for the next round of technological revolution. We hope to go together with the industry and prosper together. I hope the Kunpeng industry can have a successful future. All the best to you. Thank you.
H
Hu Guot1:48:45
Thank you. This is my third time to stand on the stage with Mr. Wu to jointly promote the ecosystem of the Kunpeng computing industry. Warm congratulations to Mr. Wu once again. Intelligent development is the biggest driving force for computing demand. Computing has now entered into a golden age. This means a huge, brand new opportunity and challenges for the industry. Digital transformation in different industries means the demand will be more diversified. In addition to general-purpose processors, there are different other specialized processors, such as GPUs and different tailor-made chips. This has also further driven the diversification of the computing industry. Green Computing Consortium aggregates the strong resources of the industrial chain. We use the open technology of ARM as a platform, and we built a neutral, open international industry consortium. We hope that we can work together with our partners to contribute to the green computing industry. Huawei is the board member of Green Computing Consortium. Huawei is always in the leading position of this industry, making tremendous contribution to this industry. We're happy to see that Kunpeng ecology has pooled the strengths and resources from different sectors, providing more choices to the partners in the ecosystem. We work together; we are open to work together with Huawei, with Kunpeng, so that we can all contribute to a healthy, prosperous, green computing industry in the world. Thank you.