Intel Spends Time at the Edge

In mid-September Intel used its Industrial Summit 2020 to announce two new chips and some new support services for edge computing in an industrial setting. Although the level of marketing for the new silicon and services was fairly low in comparison with, say, the hype machine that surrounds Apple’s new 5G iPhones, the news was quietly important in the world of industrial edge computing.

The new products take this realm, together with streaming analytics at the edge, to the next level in terms of capability. They also makes it easier and quicker for customers to develop industrial edge systems and deploy them on a large scale.

The two processors are the Atom x6000E series and the 11th Generation Core family, covering lower-power and higher-performance parts of the market respectively. Both offer significant jumps in performance on their previous versions, with faster multithreaded processing, increased graphics capability and enhanced input/output options.

But the really significant move for both is their support for time-sensitive networking and a concept that Intel calls Time Coordinated Computing.

With many industrial processes, control systems have to carry out their functions in a very short and entirely predictable time, because the results can be sensitive to small variations caused by jitter or other conditions. This is self-evident for industrial processes such as oil refining and chemical manufacturing, where additives and impurities are measured in parts per million. But it’s also true in a much wider range of manufacturing systems that produce items with precise tolerances, such as bearings, gears, sheets of metal or plastic, fibres and wires, coatings, the thickness of a paint layer and so on.

For these, the guiding principle is that the output of the process is measured and any deviation is fed back into the machine to tune the process. The feedback loop needs to do this within a certain time or the process can become unstable, creating a race condition in which the adjustment systematically under- or overcompensates because it is not getting the information it needs in time.

Time-sensitive networking and Time Coordinated Computing allow parameters to be set for networking and computing tasks to ensure the appropriate timing is maintained. Time-sensitive networking is subject to a set of IEEE standards and has been available in industrial controllers for many years, but is not widely available in general-purpose processors using high-level operating systems such as Linux or Windows.

There is a strong push within industrial edge computing to use more general-purpose computing equipment in an operational environment — effectively commercial off-the-shelf hardware rather than specialist and often proprietary systems with accompanying high prices and supplier lock-in. This is the first reason why the announcements from Intel are significant.

The second reason is that industrial edge computing is ushering in a whole range of new computing workloads built on using machine-learning models to analyse incoming streams of data. Machine learning is now used for computer vision to inspect the quality of manufactured components or fixings such as welds, for time-series signature analysis in predictive maintenance, for monitoring the audio and vibration patterns of a machine to understand potential problems, and so on.

This is a whole new area of computing workloads for industrial control systems, and Intel is — we believe — the first to provide guaranteed timing for machine learning inference on “IT grade” hardware.

To help ensure broad and rapid adoption of the new processors, Intel also said that more than 100 of its partners are aligned to the announcements, and over 200 companies have early access to the hardware to start building the processors into their systems.

Beyond the computing itself, there are other aspects to edge computing in an industrial setting, such as the diversity of machinery and use cases bringing a high level of complexity. To help address this, Intel launched Edge Software Hub, which aims to become a one-stop shop for preconfigured and tested software stacks for specific edge computing tasks. These packages of software can be downloaded and used directly by customers, or can form the basis of a custom solution. Either way, the majority of the development and configuration work will already have been done, so the task is much quicker.

This approach draws heavily on learnings from Intel’s IoT Market Ready Solutions and RFP Ready Kits for systems integrators and installers. These are preconfigured systems for specific uses that can be ordered from a supplier and put into service straight away or incorporated into a supplier system in the case of RFP Ready Kits. Intel has over 445 of these available, developed in conjunction with its IoT partners. It is working to populate a “chessboard”, made up of leading sectors and leading uses (see Intel’s Systematic Effort to Scale Industrial IoT). This pioneering approach works well and is now being followed by other players in the market.