Enterprises are using AI in conjunction with network connectivity to automate operations and generate analytical insights that improve products and services. Many are now adding 5G to connect devices to these systems because its high-speed, high-capacity, and ultra-reliable low latency characteristics can enable new use cases for AI or enhance current offerings. Until now, AI and 5G infrastructure have operated on two systems that were built and managed separately. A better and more efficient approach, however, is to integrate AI and 5G on a converged platform that can be deployed on the enterprise premises.
This article introduces the opportunity to converge AI and 5G at the edge on a single platform. It uses examples drawn from NVIDIA’s customers, products, and solutions to illustrate the role AI plays today in enterprises from the manufacturing, automotive, and smart spaces sectors. The paper describes how enterprises can further advance and improve their applications by adding 5G to their solutions at the edge via an AI-on-5G converged technology stack. The article suggests steps enterprises can follow to start taking advantage of this new approach.
Content Summary
Executive Summary
AI and 5G: Two Pillars of the Modern Enterprise
5G as the Connectivity Fabric for Enterprises
Harnessing Edge AI and 5G on a Single, Converged Platform
Focus on Manufacturing: AI-on-5G for Plant Automation, Monitoring and Inspections
Focus on Automotive Systems: AI-on-5G for Toll Road and Vehicle Telemetry Applications
Focus on Smart Spaces: AI-on-5G for Retail, Smart City and Supply Chain Applications
EndtoEnd Architecture for AI-on-5G
Build your Enterprise for the Future with AI-on-5G
Executive Summary
Enterprises are using AI in conjunction with network connectivity to automate operations and generate analytical insights that improve products and services. Many are now adding 5G to connect devices to these systems because its highspeed, highcapacity, and ultrareliable low latency characteristics can enable new use cases for AI or enhance current offerings. Until now, AI and 5G infrastructure have operated on two systems that were built and managed separately. A better and more efficient approach, however, is to integrate AI and 5G on a converged platform that can be deployed on the enterprise premises. This is NVIDIA AI-on-5G. Converging the technologies simplifies system deployment and management and gives the enterprise more control over the infrastructure. The converged approach also makes better use of computing power and the network to increase efficiencies while lowering the total cost of ownership (TCO).
This paper introduces the opportunity to converge AI and 5G at the edge on a single platform to realize these benefits. It uses examples drawn from NVIDIA’s customers, products, and solutions to illustrate the role AI plays today in enterprises from the manufacturing, automotive, and smart spaces sectors. The paper describes how enterprises can further advance and improve their applications by adding 5G to their solutions at the edge via an AI-on-5G converged technology stack. The paper suggests steps enterprises can follow to start taking advantage of this new approach.
AI and 5G: Two Pillars of the Modern Enterprise
Edge AI and 5G are becoming two pillars of the modern enterprise, as organizations adopt both technologies to drive digital transformation and reshape how they do business.
Enterprises are already using AI to provide the brains of automation and IoT, and many are running their AI applications at the edge to gain the benefits of processing data closer to where it is generated, captured, and used. Edge AI enables computer visionbased robotics systems to visually inspect hazardous places or monitor production lines; makes it possible to assimilate and interpret millions of visual images for automotive system applications; and facilitates the creation of “smart spaces” that automate retail shopping or optimize traffic and parking in cities, among other use cases. The applications, hosted at the edge, can also lower latency for time-sensitive applications and reduce transport and backhaul costs for dataintensive applications.
5G brings complementary, strategic capabilities to edge AI systems, providing the underlying connectivity for billions of devices and extending the reach of AI algorithms and applications to all connected objects at the edge. Designed specifically to serve vertical industries, 5G offers extremely fast data speeds, high capacity, and ultrareliable, low-latency communications needed for the most demanding applications. Because of these and other features, 52% of enterprises consider 5G very important to the success of IoT, according to GSMA Intelligence.
5G as the Connectivity Fabric for Enterprises
5G brings operational practicalities to the enterprise. 5G runs on a cloudnative, virtualized, and software-defined architecture that facilitates edge deployments. It can be combined with multiaccess edge computing (MEC) platforms to deliver AI, IoT, video analytics, and other strategic applications at the network’s edge. The combination of edge computing and 5G enables the ultralow latency performance needed for demanding applications. It also gives enterprises the ability to retain data onsite when edge computing resources are deployed locally to the enterprise.
5G’s many performances and deployment benefits give it advantages over earlier cellular standards and WiFi. 5G exceeds 4G by offering peak data rates of 10 Gb/s (ten times faster than 4G), latency as low as 1 millisecond (10 times lower than 4G), and the ability to support up to 1 million devices per km2 (up to 100 times more than 4G).
Connection density is also a significant distinction compared to WiFi. According to GSMA, WiFi can support just 256 to 1,024 devices on an access point with a range of 100 meters or less, depending on the WiFi version used.
The logistics of network deployments can be streamlined with 5G to increase business agility. The scale of this will vary broadly depending on the particular scenario and in comparison with other wired and wireless connectivity options. In one scenario for setting up connectivity at a new location, reported by Nemertes Research, cellular networks can be turned on in 26 minutes on average, as soon as the router is in place. An installation that relies on wired network services, by comparison, requires an average wait of 35 days, the firm reports.
Also, 5G can be implemented by enterprises as a private network. In this context, the network infrastructure is used exclusively by devices authorized by the enterprise in locations owned or occupied by the organization as an onpremise deployment. Enterprises principally choose to deploy a private network to gain more control over the network, isolate it from the public network, meet requirements for higher availability, enhanced security, and lower latency, and comply with requirements for keeping data on-premises. A private network can be implemented on a licensed or unlicensed nonoperator spectrum, which will be owned, deployed, operated, and managed by the enterprise. It can also be deployed on the operator spectrum; in this case the enterprise partners with an operator to use the spectrum through a lease or managed services arrangement.
Harnessing Edge AI and 5G on a Single, Converged Platform
The trajectory of industrial transformation for the next decade will include AI and 5G as the high-performance connectivity fabric. Enterprises around the world are embarking on this journey without a clear blueprint on the most efficient way to deploy 5G and integrate it with their AI applications. Typically, 5G and edge AI infrastructures are evaluated, designed, procured, deployed, and managed separately. The siloed approach is inherently an inefficient duplication because both AI and 5G run on computational power that can be provided and optimized by the same platform.
Fortunately, technology advancements now make it possible to deploy AI and 5G on the same computing infrastructure, enabling edge AI to run seamlessly over 5G. This is the NVIDIA AI-on-5G platform approach. Enterprises can implement the capability with a converged, high-performance “data center in a box” that is deployed onpremises and managed by the enterprise to support all workloads for AI and the 5G. The 5G workload is implemented as an additional software stack on the platform.
For enterprises, this means a single onpremises computing infrastructure can support both softwaredefined 5G as well as AI applications for computer vision, factory automation, immersive reality, and many others. Convergence bolsters the performance of existing AI applications and unlocks new capabilities for all industry segments while delivering technical efficiencies and reducing the TCO for equipment, power, space, and other resources. Converging AI and 5G in one platform also removes many of the barriers IT and operations teams normally face when trying to integrate AI with 5G for enterprise applications over a private network.
Focus on Manufacturing: AI-on-5G for Plant Automation, Monitoring and Inspections
In the manufacturing sector, AI-on-5G can make production, monitoring, and inspections more intelligent, autonomous, and instant. The following examples from NVIDIA’s customers illustrate typical use cases and benefits.
Production line monitoring for automotive manufacturers
Automotive manufacturers are using AI to visually monitor assembly lines and generate analytics insights that increase the reliability of their production lines and processes. Many are enabling the capability with a realtime video analytics system, developed by a leading computer vision software firm, which runs on an NVIDIA Metropolis AI application framework for computer vision. The visual intelligence improves the speed and precision of part picking and assembly and detects noncompliance.
Another car manufacturing company is using several small deep learning-based cameras, trained on NVIDIA GPUs installed directly in sheet metal presses, to find hardtodetect cracks in the sheet metal in a matter of seconds. The application is being used, in addition to visual inspections performed by employees, to optimize manufacturing quality inspections. 5G connectivity would give these manufacturers the capability to securely manage, monitor, and control their operations with ultralow latency. In addition, it would provide an alternative to wired connectivity, enabling flexibility and easier, faster reconfiguration of the factory floor. It could also support imaging from greater numbers of devices and greater density of devices in a location to improve defect detection. Converging AI and 5G on one platform is highly desirable for these enterprises because it can help overcome space constraints that may exist. It can also deliver operational efficiencies in a manufacturing environment with stringent unit economics.
Robotic inspections at a petrochemical plant
A petrochemical company is using AI computing at its processing plants to enable heattolerant, autonomous robots to inspect equipment and facilities under dangerous conditions with infrared cameras, chemicals, and other sensing technologies. The robotics, inspections, neural networking, and advanced data analytics were developed by an industrial robotics platform provider on an NVIDIAcertified system for the data center and can be extended to edge locations.
In such a sprawling and safetycritical environment, 5G provides a wide-area coverage for a highthroughput, lowlatency connectivity to support the robots. It would also give the plant managers the ability to support a greater density of robotic equipment and sensors at each plant and facilitate realtime controls to fully optimize the safety and productivity of the plant. Integrating AI and 5G on a single computing infrastructure would streamline deployment and management of the computational and connectivity infrastructure for the plant for improved operations and lower TCO.
Computer vision analysis of warehouse shipments
A large North American bottling company has implemented an automated computer visionbased system, developed by an AI application provider, to make sure products picked for shipping match the shipment records. It also issues alerts when anomalies are detected. The AI solution was built on an NVIDIA Metropolis AI application framework and runs on NVIDIA supercomputers for AI that are deployed across the warehouse.
In an operating environment with a lot of mobileconnected devices and equipment, including automated guided vehicles (AGVs), 5G connectivity would provide flexibility, high data rates, ultralow latency, and high traffic capacity to instantly detect errors from high volumes of sensors. Combined deployment of AI and 5G on the same infrastructure would deliver performance gains to support the increased operational efficiencies of the warehouse.
Focus on Automotive Systems: AI-on-5G for Toll Road and Vehicle Telemetry Applications
AI is bringing automation and new analytics insights to organizations throughout the automotive ecosystem. Examples from NVIDIA’s customers show how AI-on-5G can bring new capabilities to vehicle telemetry and toll road applications.
Autoracing telemetry
The motorsport arm of a leading automotive manufacturer is using AI, powered by an NVIDIA AI server, to optimize the performance of its racing vehicles during competitive events. Before each race, the company trains the AI model to recognize all cars in the field. During the race, as cars drive around the track, the AI analyzes video feeds and assesses the team’s car’s performance. The race teams receive immediate insights and adjust their cars to stay ahead of the competition each time their cars enter the garage for refueling and maintenance.
5G’s high bandwidth, ultrafast data speeds, and ultralow latency enable instant delivery of highresolution video images from the vehicles. Convergence with AI-on-5G would streamline the infrastructure on the circuit, helping to improve performance in an industry that operates on the cutting edge of efficiency, reliability, and safety.
Automation of toll road systems
Two examples illustrate the value of AI-on-5G in toll road applications. In one implementation, a major roadway infrastructure organization uses an automated tolling system to collect tolls from 900 million vehicles that cross through 1,200 tolling stations every year. The automated system also identifies each type of vehicle to estimate roadway wear and manage roadway maintenance to reduce operating costs. The company’s inhouse AI software firm developed the automated system using AIenabled realtime video analytics powered by an NVIDIA edge computing appliance at each toll station.
In another implementation, a global vendor of automatic license plate reader (ALPR) technology provides vehicle detection and classification algorithms that are used by toll road operators to read license plates as cars pass through their stations. The ALPR solution uses an NVIDIA platform that empowers cameras to process and analyze video in real time.
5G can enable faster and more accurate automation in these settings because it supports high-resolution imaging, ultrafast data transmission rates, and ultralow latency connectivity. Converging AI-on-5G would streamline the deployment of new technological capabilities for connected road infrastructure and connected vehicles to improve road efficiency and safety.
Focus on Smart Spaces: AI-on-5G for Retail, Smart City and Supply Chain Applications
Organizations across industries are using AI and computer vision applications to transform physical settings into smart spaces that offer improved productivity, efficiencies, and safety. The following examples from NVIDIA’s customers illustrate the benefits AI-on-5G can bring to retail, smart city, and supply chain applications.
Smart retail
In the retail industry, an AI solution provider is adding AI to shopping carts to streamline instore shopping and checkout. The solution uses an NVIDIA AI server and edge computing module to process realtime video analytics. The solution helps consumers navigate the supermarket, delivers personalized offers to the consumers as they shop, and enables digital payments.
Retailers are looking to 5G to deliver an enhanced, futuristic, and secure shopping experience. 5G can deliver data from high volumes of vision sensors and other devices at ultrafast speeds to deliver real-time insights that improve the consumer experience and store performance. Converging AI-on-5G would deliver a connected “data center in a box” faster, enabling retailers to accelerate the transformation of their outlets with the new capabilities.
Smart city intelligent video analysis
Cities are using AI in public settings, school campuses, and shopping centers to improve services, efficiencies, and safety.
In one implementation in East Asia, AI is accelerating intelligent video analysis (IVA) of visual data from millions of cameras deployed in cities for security, people counting, and other applications. The service uses deep learning and intensive computing power provided by an NVIDIA computational engine for AI. Another solution, based on an NVIDIA edge AI application framework and deployed in the United States, uses AI to collect and analyze multiple streams of video data to improve traffic flow, enhance pedestrian safety, and optimize parking.
5G can deliver ultrareliable, low-latency data from thousands of cameras and other devices placed at various distances around the cities, to dramatically improve the performance and efficiency of these systems. Converging AI-on-5G would streamline infrastructure deployment for city planners, delivering operational improvements for city managers and lower TCO for taxpayers.
Smart supply chain locations
An AI video analytics solutions provider is deploying its solution at distribution and supply chain facilities to help reduce the risk of COVID19 infections among workers on site. The solution uses warehouse security cameras and an NVIDIA edge AI application framework to identify social distancing conditions that might increase the risk of infections.
5G’s extremely fast data speeds and high bandwidth can ensure at-risk conditions are avoided to help protect worker safety and wellness. Converging AI-on-5G with on-prem technology would provide instant analytics needed for this solution without having to backhaul traffic to a cloudbased analytics engine.
End-to-End Architecture for AI-on-5G
The endtoend architecture for AI-on-5G includes four building blocks that can be deployed and managed by the enterprise. The building blocks include:
- Enddevices built with 5G connectivity
- 5G ORANcompliant split 7.2 radio unit that connects to the AI-on-5G server via a standard eCPRI fronthaul interface
- AI-on-5G hardware, built for the edge, with commercialofftheshelf servers running NVIDIA ConnectX® SmartNICs— NVIDIA’s latest converged accelerator that combines the powerful performance of the NVIDIA Ampere GPU architecture with the enhanced security and latencyreduction capabilities of the NVIDIA BlueField®2 data processing unit (DPU) and partners’ CPUs
- A converged AI-on-5G softwaredefined computing platform that supports the entire software stack needed to implement 5G base station and 5G networking functionalities, as well as a separate software stack needed to support edge AI services for the enterprise, industrial, and other applications.
Build your Enterprise for the Future with AI-on-5G
NVIDIA AI-on-5G is all about building a highperforming enterprise for the future. It is a new technology implementation designed for enterprises that want to use both AI and 5G to improve operational performance and want the operational flexibility and system control that an onpremise solution brings. Enterprises can build new solutions on the platform or port applications that already run on NVIDIA’s application framework to the converged platform.
The best approach when implementing a new solution is to start small, learn how to use it and how it can fit into existing operations, and scale it in incremental steps. This is also the recommended procedure for getting started with AI-on-5G. For example, an enterprise can deploy the AI-on-5G hardware and software platform with one 5G camera along with an AI application that analyzes the video feed. Once the setup is complete, the enterprise can scale the deployment at its own pace to support more cameras and 5G radios or applications.
NVIDIA’s AI-on-5G solution is offered through NVIDIA’s sales partners who engage with enterprises to understand their requirements and help them identify the appropriate configurations for their needs.
Customers can get started with the AI-on-5G Enterprise Kit, a quick launch pack made up of a 5G connected camera, a 5G radio unit and the AI-on-5G unit built on NVIDIAcertified servers. The components will enable any customer to quickly set up and explore how to use AI-on-5G for intelligent video analytics leveraging NVIDIA’s AI enterprise platform.
