Manufacturing the Future with AI and 5G

Companies can lead by example, just like human leaders. I’ve recently been made aware of a notable case study of using AI and 5G to produce AI and 5G products. Essentially this is a company “dogfooding” — using its own products to become more efficient and allowing the company to learn from direct experience of those products, which then informs future improvements. It’s common for software teams to use beta versions to spot bugs early and fix them, but using finished products internally should become more widespread, as ZTE is doing.

As a veteran — read old — analyst, I’ve seen many examples of the effect of dogfooding and employees’ personal experience in the creation of consumer products. Two examples stand out. I vividly recall standing in dusk light in the middle of a December day in a Nordic city and belatedly realizing why a smartphone-maker had launched a smartphone camera with optical image stabilization years ahead of most of its competition. Why? Such stabilization enables longer shutter speeds and hence better photos in hours of semi-darkness that are normal during daytime at those latitudes in winter.

A few years later, I tried on a $10,000 watch at a technology company’s launch event, its features focused on health. Why? The previous CEO had had long-standing health problems and by then had passed. His successor was clearly personally invested in health and fitness. As a good analyst, I can’t prove a connection and can only highlight a correlation, but I’d be amazed if the events were unrelated. And, as a visible user of the watch, this CEO will naturally have an additional level of understanding when reviewing and making decisions about watch product road maps.

Although similar examples may be less obvious in network and technology manufacturing, the benefits are the same: companies gain a greater depth of understanding from direct use of their own products and have added reasons to ensure they work effectively and to build a road map that prioritizes the most important enhancements. Customer councils provide a similar function, but the feedback loop is longer, so they are a complement and not a replacement.

Smart Learnings from a Smart Factory

ZTE’s smart factory in Nanjing Binjiang uses AI and 5G to make servers and equipment that enable AI and 5G. It offers ZTE a “virtuous circle” for incremental improvements, as well as leading to near-term efficiency benefits. ZTE reports that its use of 5G-connected machine vision for quality inspection cuts the miss rate by 50%. The company claims that a combination of AI computing, robotics and machine vision result in a 20% improvement in assembly quality.

To measure the energy efficiency of its products, ZTE uses AI and 5G to run more than 100 low-power tests for nearly 1,000 cycles. ZTE points to other benefits from increased automation, including a 40% boost to the efficiency of packaging.

5G Enables Greater Factory Automation through Autonomous Vehicles

Unlike Wi-Fi, the use of 5G enables a deterministic network where performance can be guaranteed to be sufficient for autonomous guided vehicles (AGVs) to move goods around the factory. As AGVs are large, their movement must be safely managed. In ZTE’s factory a single assembly line can make multiple types of server. Each server can be built to a different specification, which complicates the production process. Close management of AGVs is needed to ensure the routes they take are as efficient as possible.

Devices across the factory are connected by 5G to an intelligent operations centre that manages the entire process. Algorithms drive factory management decisions using data collected in the factory and data on production requirements. This integrated approach also minimizes energy usage, helping ZTE to achieve its sustainability and cost targets. As a result, the company reports that efficiency has been boosted by 125% and delivery times have dropped 48% compared with traditional production methods.

Managing the Heat Created by AI Servers

Air cooling is reaching the limits of its usefulness to control the heat created by servers efficiently. AI workloads are placing increasing thermal load on servers, just as companies now have many new sustainability targets. This need for greater computing power carries the risk of much higher energy costs for cooling and increased carbon emissions unless more-efficient cooling solutions are deployed.

Liquid cooling offers many energy-efficiency advantages for data centres over fans. ZTE has two approaches that aim to offer low ratings for power usage effectiveness (PUE). By placing a cold-plate cooling system in contact with the CPU, GPU or memory, ZTE reports achieving a PUE as low as 1.1, while supporting up to 400 W on a single chip. This approach retains easy access to components for upgrades and maintenance.

For greater performance, ZTE has an immersion cooling solution that is commercially available. For high-energy workloads such as AI, where maximum energy efficiency is needed, or where data centres have limited space, for example in urban locations, an immersive solution is ideal. ZTE reports its immersive liquid cooling offers a PUE of less than 1.09 and supports cooling for over 2,000 W of heat using indium sheets. The company is choosing not to disclose the precise composition of the liquids it uses as it tries to differentiate from competitors, but it supports both oil-based and fluorinated liquids.

Navigating GPU Supply Problems

The supply of the GPUs and the computing power needed for AI is increasingly uncertain. Supply chain rules often change at short notice. And cryptocurrency users can swallow large quantities of AI hardware suddenly if the balance between the cost of buying and running hardware and the economic return of currency mining turns favourable.

Yet AI workloads are becoming more important to support an ever-wider range of products. Plus, the workloads are becoming bigger as usage increases and model sizes balloon. So, AI hardware customers need both extensive compute capability and a predictable supply of such powerful AI hardware.

To work within these supply challenges, ZTE’s servers feature an “open and decoupled” architecture, designed for flexibility and compatibility with multiple technologies. This enables the use of a wide range of GPU cards, including highest-performance and cost-efficient GPUs from diverse vendors.

Flexibility applies to other components too, such as memory, and support for the latest Intel CPUs such as Xeon 6 Efficiency Core processors. ZTE works with more than 500 suppliers to maximize supply chain stability and speed innovation.

AI also increases the size and performance requirements of data flows, and ZTE believes its expertise in network interface cards (NICs), storage acceleration and transmission networks complements its AI servers to offer the most effective overall solution for AI workloads. ZTE’s smart NICs aim to boost forwarding performance by 200% while reducing power consumption by 41%; they connect to ZTE’s 400GE and 800GE high-throughput interconnections to link AI servers.

ZTE Aims to Boost Its Leadership Credentials

By communicating about the way it is using AI and 5G in its 5G factory, ZTE aspires to become an innovation leader. Perhaps more importantly, the company’s use of its own servers, 5G and AI products in its factory will help ZTE develop the next generation of products more effectively.