Specialized processors are changing how computing performance improves. For industrial automation, this shift matters because control, analytics, machine vision and other workloads do not all place the same demands on hardware. A processor tailored to a particular calculation can perform it faster or with less energy, but it may be less flexible than a general-purpose processor. As more computing investment moves toward specialized designs, improvements may also become less evenly shared across applications.
This tension is the central argument in a 2018 analysis of computing as a general-purpose technology. Its concern is not that computers will lose the ability to run certain programs. Rather, computing platforms could fragment: some workloads receive rapidly advancing, customized hardware, while others rely on general-purpose processors whose performance gains are slowing.
Why general-purpose computing became the norm
A general-purpose processor can run many kinds of software without requiring its hardware to be redesigned for each task. This flexibility helped make computing useful across a wide range of industries and applications. As demand expanded, chip manufacturers had a larger market to support investment in better processors. Improved performance, in turn, made computing attractive to more users. The paper describes this reinforcing relationship as the virtuous cycle of a general-purpose technology.
For decades, improvements in processor performance per dollar helped sustain that cycle. When a broadly useful processor improved quickly, organizations had less reason to fund specialized hardware: the next generation of general-purpose chips could soon narrow the performance gap, while a custom design carried substantial development costs and offered less flexibility.
That balance began to change as improvements in general-purpose processors slowed and the cost of advanced chip design and manufacturing rose. The paper argues that when broad-platform gains become less rapid, the performance advantage of a specialized processor can remain valuable for longer. That changes the economics of choosing between flexibility and workload-specific efficiency.
What specialized processors trade for performance
General-purpose designs must accommodate many types of calculations. Specialized hardware can instead make design choices around a narrower workload. Those choices can affect how much computation happens in parallel, how data moves to the processor, and how much energy is used. The benefits depend on the workload rather than applying equally to every application.
For example, graphics processing units (GPUs) can perform many calculations in parallel, making them well suited to tasks that can be divided into independent operations. Their memory systems also favor workloads where data needs can be anticipated. By contrast, a general-purpose central processing unit (CPU) is designed to handle a broader range of tasks, including those that do not map neatly to parallel execution.
This distinction is relevant when evaluating automation workloads. Image analysis or other computation-intensive tasks may have different hardware requirements from software that must accommodate varied operations. The appropriate choice depends on how the work is structured, how predictable its data access is, and whether its calculations can be parallelized. Specialization is not a universal upgrade; it is a design trade-off.
Energy efficiency is another reason to consider specialized hardware. The paper highlights its value for battery-constrained mobile and Internet of Things devices, as well as for data centers where computing is performed at large scale. In industrial settings, power constraints may be part of the hardware decision, but the source does not suggest that every automation application benefits equally from specialization.
Deep learning accelerated the shift
Deep learning helped demonstrate what can happen when an algorithm and a processor architecture are a good match. Neural-network calculations can make extensive use of parallel processing. The paper describes how GPUs enabled larger deep-learning models to be trained more effectively than on general-purpose processors, contributing to the technique’s success in image recognition and other applications.
The demand generated by deep learning also encouraged development of more narrowly tailored hardware. Application-specific integrated circuits (ASICs), for instance, are designed for particular functions. Their specialization can provide a substantial advantage for the intended workload, but it also limits the range of tasks they can perform. Building custom chips involves high fixed costs, making the size and value of the target workload important considerations.
Specialized hardware also brings practical challenges beyond chip design. Software may need to be adapted, and programming can be more difficult than on familiar general-purpose platforms. The paper points to these factors, along with development expense, as reasons specialized processors did not replace general-purpose chips across the board.
How computing fragmentation could affect automation
The paper’s main concern is that the economics of computing may form a cycle in the opposite direction. If general-purpose performance gains slow, users may move workloads to specialized processors. That can reduce demand for the broad platform, making it harder to finance further improvement. As investment and adoption shift, different areas of computing may advance at different rates.
For automation decision-makers, this argument offers a useful lens for assessing hardware choices:
- Match hardware to the workload. Determine whether the required calculations are parallel, predictable and suited to a specialized design.
- Account for software effort. Consider programming complexity and compatibility alongside processor performance.
- Weigh fixed costs against the application. Custom hardware is more attractive when the workload can justify its development expense.
- Consider the broader platform. Specialized acceleration may benefit one application without delivering the same gains to other software that depends on general-purpose computing.
These considerations do not point to a single processor strategy for every plant or system integrator. They underline the importance of evaluating the workload, the cost of adapting software, and the degree of flexibility required before prioritizing a specialized solution.
The practical implication
Specialized processors can unlock meaningful gains when hardware and computation are well matched. At the same time, the shift away from a widely shared general-purpose platform may make computing progress less uniform. For industrial automation leaders, the practical takeaway is to treat processor selection as a workload and investment decision—not simply a search for the fastest chip. Balancing performance, energy use, programmability and flexibility can help ensure that hardware choices serve the full needs of an automation system.