In 2016, at the annual meeting of the Gartner data center, Peter Levine, a Google venture capitalist, thought edge computing would be cloud computing.
Now it seems that this radical conclusion has not been confirmed, but that the business pattern of cloud, edge and end collaboration is gradually taking shape.
Edge calculation as between the ends of the cloud
Among them, Intel especially emphasizes the strategic layout in the field of edge computing.
Of course, Intel didn't lock the oasis of edge computing directly at first,
In the specific industrial application process, especially in the current network environment, Intel quickly focused its strategy on edge computing, and put forward three clear strategic directions of the Internet of things at the China Internet of things president Summit (Hangzhou) in November 2017: high performance chip, enhanced edge computing, computer vision.
At the summit, Intel demonstrated its deployment in the field of edge computing, including four versions of the OpenVINO tool suite a year, Core Rui, iterations of the Movidius hardware platform system for the strong series of chips, the configuration of multi-application fusion technology, the introduction of edge artificial intelligence edge computing ecological think tanks, and the application of machine vision in areas such as retail, industry, cities, and medicine.
On July 27 last year, Intel officially launched the openvino visual reasoning and neural network optimization tool suite at the visual solution and Strategy Conference. Since its launch, this tool suite has been updated at the rate of a new version every quarter, which has obviously become an important deployment of Intel in edge computing, especially in the field of vision.
At the summit, Intel officially released the version of openvino R3.
According to the official data, this version update is mainly aimed at the problems that openvino version occupied too much resource space, some models need to change the image size between different reasoning operations, resulting in reduced accuracy, and it is difficult to trace the performance and accuracy impact of each change on the model.
Specifically, the updated version of openvino R3 optimizes the network model loading, adds three visual pre training algorithm models, adds command-line interface (CLI) deployment manager, and supports the 10th generation intel core processor (ice lake).
At the conference, Thomas neubert, director of video business market development of Intel, stressed that openvino has been open-source based on Apache 2.0 license. As of September 13, 2019, this in-depth learning network architecture has been supported by 369 in-depth learning models.
As for the product iteration of openvino, Thomas neubert gives a set of data:
Based on the first generation of Xeon processors, openvino's R1 version in 2018 can double the computing performance, and the R1 version in 2019 can increase the xi'neng by 2.1 times. If we use the second generation Xeon processor based on the R1 version of openvino in 2019, the computing performance will be improved by 28.4x.
Manufacturers in the industry will provide raw performance data. In fact, in our use cases, raw performance data is not the most important. What we want to see is that it can bring overall performance level improvement in the end-to-end solution.
At present, openvino has continuously iterated over six versions. Based on the existing hardware platform, the computing performance has been improved from 250 frames / s to 2600 frames / s, as shown in the following figure.
In addition, openvino, as the current Intel strategic software product, has also been supported by Intel's full range of CPU products, including Intel's smart, core and Xeon series processors, movidious Vpu, FPGA acceleration chips, and future acceleration chips.
These are Intel's version iteration and system construction for machine vision and openvino, and also Intel's entry point in the age of aiot.
Why does Intel pay attention to edge computing, especially machine vision?
Lei Feng found the answer from two sets of data provided by Zhang Yu, chief technology officer of Intel's Internet of things business unit in China.
According to relevant forecasts, about 55% of the data generated by human beings in the world in 2025 will come from the Internet of things, 50% of which need to be processed on the edge, and the other half will be processed in the cloud, so edge collaboration will be the general trend of the whole Internet of things in the future.
In many Internet of things applications, we see that a large part of applications are related to video, and video data grows rapidly. According to the prediction of the industry consulting company, the growth rate of video data will reach 11% every year. It is estimated that 82% of the network data will be related to video by 2021.
Among them, there is an important concept "software as a service". According to Lei feng.com, "software as a service" is more a concept put forward by various cloud service providers, and it is also a basic idea for Intel to further expand openvino development tools. Among them, the key technology of software as a service is multi application integration.
Platform software can realize the unified allocation of hardware resources, but also can provide a standardized middle layer to break the relationship between the upper application and the lower hardware resources, so as to improve the efficiency of system utilization. We see the rapid development of these two aspects. Including software definition network (SDN), network function Virtualization (nfv), software definition storage (SDS) and other software definition system application cases that integrate several applications on a unified platform through virtualization technology are constantly rich.
Because of this, Intel is developing multi application fusion software. According to Zhang Yu, the software includes two aspects, one is focused on the edge side, the other is focused on the cloud.
On the edge side, Intel provides extensions to its existing software tools (such as openvino extension), which makes these tools better support multi application environment. At the same time, on the edge side, it also provides some detection tools to help developers detect the current working status of each application and the utilization of system resources.
In the cloud, Intel provides the reference practice of software management platform, users can input the system parameters of the fusion system on the web. After these parameters are set, the installation script will be automatically generated in the cloud management platform, and developers can download the script to run locally, so as to quickly build the required multi application running environment.
The Internet era continues to be a thinking of ecological construction. Major manufacturers attach great importance to ecology when deploying the industry for the Internet of things.
Lei feng.com (public name: Lei feng.com) noted that at this summit, Intel released the AI at the edge ecological think tank, which is also Intel's specific deployment of the fourth stage of Internet of things ecology focusing on aiot.
Yu Bing, channel ecology director of Intel's Internet of things business unit in China, explained the deployment in detail at the summit:
Edge AI eco think tank includes three sections: high performance hardware and industry software formula, pre bundled developer products and tools, specific use case suite and deployable scheme.
First, high performance hardware and industry software formula. Intel has high-performance CPU, GPU, Vpu, movidius, FPGA visual accelerator, and openvino development kit for edge computing and machine vision.
Second, pre bundled developer products and tools, jwipc, uzei development tools, Intel neural computing stick, and edge to cloud integrated product solutions;
The third specific use of Anli suite and deployable solutions, including digital China rush business automation and management process visualization solution, Shenzhen Xinbu dynamic face recognition gate control solution, Jiehe digital signage management system, Shanghai Ruishi industrial machine vision development suite, etc.
As mentioned earlier, Intel's entire Internet of things ecological construction has developed to the fourth stage, focusing mainly on marginal artificial intelligence (, AIoT (AI IoT).). In view of the four stages of development history, Yu Bing also combed at the meeting:
In stage 1.0, Intel ecosystem construction mainly includes ODM, OEM manufacturers, independent software developers, cloud service manufacturers and other end-to-end solution manufacturers, which is also the most basic ecosystem expansion based on Intel's original genes.
In stage 2.0, in 2017, Intel officially proposed the industry's overall solutions and development kits, accelerated the expansion of the ecosystem, further covered the Internet of things solution aggregators, and pushed the solutions and development kits to the market through digital China, jwipc, etc.
In phase 3.0, from 2018, Intel began to focus on the deployment of industry solutions. According to the data released by the conference, so far, Intel has more than 217 solutions and 90 development kits deployed in the world; so far, 3400 projects across the globe have been deployed and applied in 10 months;
In phase 4.0, starting in 2019, Intel focuses on rapid scale deployment of aiot through developer tools and ecosystem solutions.
As the main battlefield for the giant to compete for the Internet of things, Intel integrates its original industrial chain resources. After nearly three years of industry sinking, Intel has further broadened its ecology on the Internet of things track. Through the software and hardware systems of edge computing and machine vision, it has accurately positioned its own core system in the era of the Internet of things and even the intelligent network.
According to data given by Wang Rui, vice president and general manager of Intel Marketing Group, at the conference, Intel's new ecological chain support program was further expanded in 2019.
Two years ago, Intel put forward three strategic directions of the Internet of things at the China Internet of things CEO Summit (Hangzhou) in November 2017: high performance chip, enhanced edge computing and computer vision. Among them, edge computing, as a new direction running through Intel's three strategies, has replaced the previous statement of "workload integration" and began to be infiltrated into various industries.
At the Intel Internet of things Summit the following day (October 17), Thomas Lantzsch, Senior Vice President of Intel and General Manager of the Internet of things Division, further highlighted Intel's specific strategy in the field of the Internet of things:
Such Internet of things strategy is specifically implemented in the industry. Thomas lantzsch also said that it has laid out nine subdivisions: retail, industry, smart city / video, transportation, public utilities, education, health care, automobile, financial services. Among them, security, retail, medical treatment and industry are the important application fields of computer vision and AI reasoning.
In view of the further development planning of computer vision and edge computing development tools, Thomas lantzsch also revealed that at present, openness has been developed with partners to improve network capabilities, and will continue to launch corresponding tools for functional security and real-time control in 2020.
In addition, referring to the driving force of edge computing, Thomas lantzsch believes that it covers the following four aspects:
First, low time delay, many scenarios require applications to be very low time delay, such as robots, control systems, in the factory there are such requirements;
Second, large bandwidth. For computer vision, bandwidth is still limited. In the future, the amount of data will be huge. In terms of economic cost, we can't transfer massive data to cloud computing.
Third, connectivity, which is also a challenge for many customers;
In the past two years, Intel began to emphasize edge computing and data centric strategic transformation. After two years, such strategic deployment finally no longer floats on paper and starts to cause chemical reactions in the industry. As Thomas lantzsch, senior vice president and general manager of the Internet of things business unit of Intel Corporation, mentioned at the Intel Internet of things summit the next day:
Two years ago, we could only watch the PPT demonstration for you to tell you what we have done; last year, we have seen many industry solutions demonstration, concept verification and solution prototypes; today, the off-site solutions are all on-line solutions, and there are many actual deployments in China and the world.
In terms of industry application implementation, Wang Rui gives more introduction from the market aspect:
In the industrial field, we have made clear the strategic direction and product thinking of the workload integration of "motion vision", and each partner has also launched a series of new products and new solutions.
In the field of smart retailing, we have cooperated with our customers to launch a business terminal based on transparent computing and IDV, put forward an end-to-end solution at the education end, and put forward an intelligent conference room solution to meet the needs of the financial sector, the education sector and other industries, and further expand to the new application market;
In the field of intelligent transportation, we have identified four strategic directions of high-speed rail, subway, port and expressway, deepened cooperation with customers in these four directions, and started to gradually launch solutions based on Intel product technology in four industries;
In the field of medical and health care, in addition to continuing to use artificial intelligence to enable traditional medical testing instruments, Intel also increased customer cooperation in smart hospitals, and launched a series of solutions and products for application scenarios such as smart ward, smart medical room, and smart operating room.