Single-board computers are, as the name implies, fully functional computers that have been implemented using only a single circuit board. These computers have been around since the mid 1970s. But with the advent of personal computers (PCs), fewer single-board computers were produced and their popularity declined. PCs use a main board into which daughter boards can be plugged in, either to extend memory, provide powerful graphics and rendering capabilities, or add other peripherals. Single-board computers have a fixed setup and do not necessarily provide the same flexibility to customize peripherals using daughter boards as PCs do. However, the popularity and success of smartphones has been a main driver for the development and research of efficient and feature-rich system on a chip (SoC) processors. Unlike common processors which are used in desktops, servers, and – to a large extent – laptop computers, SoCs have many components available on their die. For example, they feature integrated graphics processing units or signal processing units that are needed for operating the display on a smartphone or to turn an analog audio signal into a digital stream when doing a voice call.
These technological advances have led to a resurgence of single-board computers in recent years. Most popular among these single-board computers are the Raspberry Pi boards. There exist various versions of the Raspberry Pi boards for embedded, industrial, or home automation use cases. In the computer science department, we have used the Raspberry Pi on multiple occasions for our research as they provide reasonable computational performance and consume very little power compared to other devices at our disposal. This makes these single-board computers ideal for embedded devices used in smart homes, smart cities or even survey stations. The principal reason that SoCs are consuming less power than regular processors is because of their instruction set architecture (ISA). There exist two classes of ISAs which describe the fundamental design of processors. These are complex (CISC) and reduced instruction set computers (RISC). SoCs make generally use of RISC architectures, which require fewer transistors for implementing than CISC architectures. The requirement for more transistors in CISC architectures is due to the translation of CISC instructions to RISC instructions, hence, CISC architectures are essentially making internal use of a RISC architecture. The fewer transistors architectures need for implementing processors, the less power they will consume.
One area in which current and future single board computers could excel in are offices. In particular, the continuous performance improvements make single board computers a viable alternative for office use. In offices, computers are primarily being used for document reading and writing, browsing the web, programming and for running some tests and doing analyses. However, none of these tasks are computationally demanding and they mostly show a bursty performance behavior. Consequently, office devices are often operating in low power states, which are less energy efficient than if the full computational potential of the device was used. A simar issue occurs also in cloud data centers, where virtual machines and containers of under-utilized machines are being migrated onto different physical machines in order to raise the overall machine utilization and increase energy efficiency. Any machine that is not being used at all can be turned off to save energy. Only when processors and devices as a whole are executing at their peak performance can optimal energy efficiency be reached. This is because static power makes up for a majority of the total device power consumption. 20 years ago, the situation was different. Then, dynamic power was the dominant term in total power consumption. The shift in power consumption also led to the introduction of multicore processors, as it was still possible to shrink the size of processors but it was no longer possible to improve their frequency.
If office devices were to be replaced by single board computers, which are perfectly capable of executing the previously mentioned tasks (at the cost of taking a bit more time), then not only would the utilization of the single board computer be higher, but also the energy efficiency would be higher. This change could save lots of energy for bigger companies and organization, while devices would also be dissipating less heat. With less heat being dissipated, less energy has to be spent for cooling down devices and less noise is produced by fans. As temperatures are globally rising and in the face of more extreme heat waves during summer, being able to work in a cool and quiet environment becomes a much required need. Single board computers which are integrated into keyboards, such as the Raspberry Pi 400, are promising candidates for this kind of use case. Paired with a low-power display, as little as 30 watts could be consumed in total. Compared to typical office setups with all-in-one PCs, or even thin clients, this can make a difference ranging between 15 to 80 watts (which is similar to the heat dissipated by a light bulb). The next time you find yourself sitting in front of a PC and complaining about the temperature in the office, maybe consider changing to a single board computer setup. You might as well give it a try if your workload is suitable for single board computers and if you are willing to accept longer load and execution times. Not only is it favorable for cooler office temperatures, but it is also a more pleasant experience to be working in a quiet environment without fan noise.
Auteur(s) de cette contribution :
Doctorant à la Chaire de systèmes complexes de l'institut d'informatique à l'Université de Neuchâtel. Je recherche et développe des solutions autour de la sécurité et l'efficacité énergétique des systèmes distribués, notamment des blockchains.
PhD student at the Complex Systems research group of the Computer Science department at the University of Neuchâtel. I am researching and developing solutions around the security and energy efficiency of distributed systems, including blockchains.