AI computer
The eight treasures of the world artificial intelligence conference 2020 are respectively Microsoft silica project, the first "intelligent shelter hospital" solution in China, ABB's Yumi al arm robot, Huawei atlas 900 training cluster, Uber Walker robot, IBM artificial intelligence debate system, quadruped robot opencat, intelligent grazing robot
1. Microsoft silica project
is located in the "Ai pilot area", which makes it possible to store a large amount of data in glass. With the help of this carrier, up to 75.8g of data can be stored in a 7.5 square centimeter glass sheet
2. The first "intelligent shelter hospital" solution in China. During the COVID-19 prevention and control period, more than 30 intelligent robots such as cleaning, disinfection, security and so on demonstrated their skills in the Hubei Wuchang shelter hospital. p> 3. ABB's Yumi al arm robot is on display in the "Ai + instry" section. This time, Yumi brings intelligent medical solutions covering the whole chain of hospitals, from biotechnology, gene detection, pharmacy to logistics. The audience can also listen to the growth story of human-computer cooperation in the past five years 4. Huawei atlas 900 training cluster is located in the exhibition area of "Ai + basic technology". Atlas 900 is integrated by thousands of shengteng 910ai processors, and its computing power is equivalent to 500000 computers, representing the peak of global computing power with powerful computing power, Atlas 900 can help researchers engage in weather forecasting, oil exploration, exploring the mysteries of life, accelerating the commercial process of automatic driving, etc., which will benefit the world as "made in China" in the future 5. You must choose Walker robot you must choose Walker robot appeared in "Ai + ecation" exhibition area. The robot integrates artificial intelligence and robot core technology, with 36 high-performance servo joints and omni-directional perception systems of force sense, vision, hearing and balance, achieving breakthroughs in whole body motion control, flexible walking in complex terrain, self balance, hand eye coordination, visual recognition, intelligent and safe interaction, etc 6. IBM AI debate system will participate in the selection of sail award. Debate is a subjective art form and skill. It is human language. IBM artificial intelligence debate system is recognized as one of the major challenges in the field of AI. In the future, more people will live in social media space and fragmented time, and the system will help people make more intelligent decisions 7. Four legged robot opencat the four legged robot opencat developed by petroi, an entrepreneurial team from Silicon Valley, made its debut in China in the "Ai instry ecosystem". Small body has super brain, keen perception and activity ability, human-computer interaction, automatic balance, flexible obstacle avoidance and other abilities. In addition, bittle, petoi's second-generation proct, will also be launched in the world 8. Intelligent grazing robot intelligent grazing robot is a major special achievement of agricultural work in Tibet Autonomous Region, and this is also the first time. This proct solves the problem of cattle survival rate that herdsmen are most concerned about, including timely treatment and targeted driving of Yaks in the early stage of disease, which will comprehensively improve the economic benefits of herdsmen
The development of single CPU can not meet the needs of practical applications, and the AI era must rely on parallel computing. At present, the mainstream architecture of parallel computing is heterogeneous parallel computing platform. If you need the service of computing power, you can go to the tenth power
as we all know, the common server is the provider of computing power based on CPU, which adopts the serial architecture and is good at logical computing and floating-point computing. Because a lot of branch jump processing is needed in logic judgment, the structure of CPU is complex, and the improvement of computing power mainly depends on stacking more cores
however, with the application of network technologies such as big data, cloud computing, artificial intelligence and the Internet of things, the data in the Internet is growing exponentially, which poses a serious challenge to the traditional services with CPU as the main source of computing power. At present, the processing technology of CPU and the number of cores of a single CPU are close to the limit, but the increase of data continues, Therefore, the data processing ability of the server must be improved. Therefore, in this environment, AI server came into being<
nowadays, AI servers in the market generally adopt the form of CPU + GPU, because GPU is different from CPU, which adopts the mode of parallel computing and is good at sorting out intensive data operations, such as graphics rendering, machine learning, etc. On GPU, NVIDIA has obvious advantages. The number of single card cores of GPU can reach nearly 1000. For example, the number of cores with 16 NVIDIA Tesla V100 tensor core 32GB GPUs can exceed 10240, and the computing performance can reach 20 billion times per second. And after years of market development, it has been confirmed that CPU + GPU heterogeneous server in the current environment can really have a lot of development space
but it is undeniable that every instry needs to go through a lot of wind and rain from the beginning to maturity, and in this process of development, competition always exists, and can promote the sustainable development of the instry. AI server can be said to be a trend or a rising force, but there is still a long way to go for AI server. The above is the answer to the tenth power of Inspur server distribution platform.
on the afternoon of August 23, 2019, Huawei's Shenzhen headquarters released shengteng 910, the most powerful a processor, and launched a full scene AI computing framework Xu Zhijun, chairman of wheel of Huawei, said, Huawei has completed the construction of AI solutions for the whole station and scene, and will launch more AI processors in the future to provide more abundant, more economical and more suitable AI computing power but it gave him a good excuse. He kept shouting that the wolf was coming, the wolf was coming<
The first echelon: Kirin 970 and apple a11: independent AI processing unit and AI software system, as well as open AI underlying technology strength
the second echelon: Qualcomm snapdragon 845 has no independent NPU, and its software system is also independent. The opening of underlying technology depends on various mobile phone manufacturers, unlike Huawei and apple, which have the source of technology
the third echelon: MediaTek P60 lacks an independent NPU, and its software system relies on various companies. Although it has the blessing of oppo R15, it has always been linked with the low-end machines of many manufacturers (Hongmi, Meilan, etc.), so it is not good in momentum or strength
specifically, we can look at the gap in three aspects
Comparison of Architecture:
970 has NPU unit, while 845 is the traditional CPU + GPU + DSP architecture. Snapdragon 845 does not speed up the relevant hardware, but uses DSP to process the relevant operations. In fact, it is meaningless to directly accumulate the computing power of CPU, GPU and DSP. Because these are shared resources, it is impossible to use all of them for AI computing. At present, a network model can not be split into different computing entities. The performance of Kirin 970 only uses NPU, and does not include CPU, GPU and DSP. Compared with the traditional CPU, NPU has 25 times performance and 50 times energy efficiency advantage in the neural network algorithm. At this point, 970 has won a lot
comparison of calculation ability:
through the evaluation of AI running score, we can see that 970 vs. 845 have obvious advantages in basic calculation ability
Master Lu AI score test, Qilin 970 crushing Xiaolong 8451. Master Lu AI score:
Qilin 970 score 233, Xiaolong 845 score 190. 2. NPU is more efficient and full-time: if one word is used to describe Qilin 970's NPU, it is "specialized in technology". Only when it comes to AI task processing, it will be handed over to NPU independent chip for full-time computing processing, Moreover, under the same task, the efficiency of NPU is much higher than that of CPU + GPU + DSP distributed heterogeneous computing
Distributed heterogeneous computing: since CPU, GPU and DSP are not designed specifically for AI computing, it is not surprising that they are not as efficient as single NPU in terms of execution efficiency. The other situation that running score software does not involve: when CPU, GPU and DSP are already in full load, the mobile phone chip can not provide enough AI computing power, and the efficiency will be further greatly recedAI Ecology:
focus on showing the difference between AI and non AI. Qilin 970 hIaI has established an ecology, and new applications appear one after another. At present, Qualcomm is still on paper
specifically, mobile phones can be directly benchmarked, compared with Samsung S9, S9 + app opening, red envelope grabbing, games, etc.
so at present, there are only two companies in the world that master AI chips: apple and Huawei
when it comes to the true flagship of AI, many so-called AI mobile phones in the market are actually embodied in the form of AI experience, such as AI photography and AI voice communication (Samsung S9, Xiaomi, etc. equipped with snapdragon 845), rather than AI based performance. So these are not really AI phones
I think that if we really want to use mobile phones as the flagship of AI, it is the Qilin 970 series procts. Take glory V10 for example. The AI performance of glory V10 comes from the soft and hard AI support of AI chip Kirin 970 and emui8.0 artificial intelligence system. Qilin 970 adopts the hIaI mobile computing architecture. The AI performance of NPU is much better than that of CPU and GPU. It handles the same AI tasks and has about 50 times energy efficiency and 25 times performance advantages. This means that Kirin 970 can complete AI computing tasks faster with less energy consumption, and it also means that CPU and GPU are liberated from AI processing, achieving better results
in short, the three most outstanding and core features of glory V10 are independent AI processing unit, supporting AI operating system, and open underlying AI technology strength and ability. These three points are also the reasons why he can stand out in the same instry. So he's the true flagship of AI
However, I feel that AI chip technology is still in its infancy in mobile phone applications, and the technology has not yet reached the mature stage, and there are still many immature and imperfect places. It takes time for manufacturers to update and promote technology, and it takes time for the market to accept and digest. However, in any case, the era of AI smart phones has come, and the speed is rapid, and the AI chip technology is bound to develop rapidly and matureThe annual salaries of the two scientists reach one million dollars. It is reported that Alibaba artificial intelligence laboratory is mainly composed of R & D personnel, accounting for about 80%. In addition to Chen Ying and Tan Ping, the core proct and technology R & D team also includes Chen Lijuan, the person in charge, Nie Zaiqing, the chief scientist of voice technology, Li Jianye, the chief designer, Ru Yi, the general manager of hardware terminal, Du Haitao, the general manager of proct operation, etc
According to Ali, Dr. Chen Ying's professional accumulation can help Ali improve its visual ability and better empower aiot ecological partners. At the same time, Ali's business can provide rich research scenarios for it to continue to explore the machine's understanding of the world and people. In Tan Ping's opinion, Ali mainly focuses on his own experience in algorithm. "In recent years, AI has been very popular, among which there is more work to do two-dimensional image recognition and detection, but there is still a lack of three-dimensional vision."“ The realistic 3D reconstruction technology that Professor Tan Ping is engaged in can help Ali build 3D models of commodities and stores and create immersive e-commerce experience. " Ali said. Tan Ping said that the team may do a holographic world project in the future. By establishing a three-dimensional digital version of the real world, users can do many things in the real world in the virtual world. For example, users can shop in the virtual store, and the store can also get rid of the space restrictions, and update the space or goods in real time, which is more rich and three-dimensional than the simple commodity list, so as to improve the shopping experience of consumersin addition to shopping in 3D space, Tan Ping and his team will also study the landing of AR or robot navigation procts in the future
Text: by
lidar, camera and other sensors are just like the eyes of automatic driving vehicles. If a car wants to realize automatic driving, it is essential to have "good eyesight". As a leading sensor supplier in the world, Scala, a proct of Valeo, is the only vehicle size lidar that has been mass proced and installed in the listed models so far. In addition to sensors, there are also a lot of car related AI applications on display at CES and Valeo. Today, I'm in Las Vegas to show you these black technologies that will be displayed at CES
predictive AI movepredict. AI
although the brain's reaction speed is far slower than AI in complex information processing and calculation, for example, human beings are always abused by AI in chess, in fact, the "computational power" of human brain is far higher than AI. The advantage of human brain's high "computational power" can be reflected in some scenes at present, human drivers have much higher ability than automatic driving vehicles, such as the "game" between pedestrians and non motor vehicles when crossing the road, the "game" between themselves and other driving vehicles in the roundabout, and human drivers have better understanding ability and coping style in identifying the driving intention of the target
understanding the heart rate, body shape, body temperature, emotion and other information of each passenger in the vehicle can help the vehicle provide personalized service to the actual needs of each passenger and make the vehicle more intelligent. The heating system of the demonstration car has also been modified. The traditional electric heating (PTC) has been changed into a heat pump which can be controlled by different zones. It can not only meet the temperature requirements of different members in the cockpit, but also improve the utilization efficiency of air conditioning
by using low-energy heat pump, unnecessary heating demand can be avoided (for example, only heating for the driver when only driving), so that the vehicle can save up to 30% of energy consumption in winter. This technology is especially suitable for pure electric vehicles, air conditioning efficiency is higher, also can rece the waste of battery energy consumption, and ultimately effectively improve the pure electric vehicle winter range
this article comes from the author of car home, which does not represent the standpoint of car home