A Comparative Study on the Power Usage of Raspberry Pi Models

B. Features under Investigation

As RQ1 and RQ2 delve into examining the power usage of key functionalities across different platforms, it was necessary to create a comprehensive list of such common operations. To achieve this, the common usage scenarios identified in previous research [8, 10] were collected and expanded upon to encompass a broader range of tasks. During this compilation phase, only tasks that could be universally performed across all platforms were included. Consequently, tasks like sending an SMS or making phone calls, which are specific to smartphones, were excluded. The compiled features being investigated are detailed in Table VI.

Various operations outlined in Table VI required the utilization of specific software. To accomplish this, the version compatible with the device being tested and its operating system was employed. Additionally, certain operations necessitated the utilization of specific files (such as .mp3 files for O9, movies for O10, and images for O12, among others), and in such cases, identical files were utilized across all platforms whenever feasible.

C. Measurement Methodology

Due to the absence of a standardized power measurement tool compatible with all platforms under examination, different hardware and software were employed to measure the power consumption of operations outlined in Table VI. When devices could be powered directly from the outlet using electric current, the Eco-Worthy electronic wattmeter was employed. Hardware-based measurement was favored over software due to its superior accuracy [10]. To use the power meter, the device was plugged into the power socket, and then the device under test was connected to the power meter. It’s worth noting that while the meter itself requires electricity to operate, it registers 0.0W when no device is connected to it.

Additionally, only CPU power was taken into account during the experiment to ensure more accurate comparative values for RQ2. For the RPi and desktop computer, CPU power could be directly measured using the power meter. However, for the laptop, the power consumed by the screen component had to be deducted from the power meter readings. To measure screen power consumption, Microsoft Joulemeter was calibrated and employed. Furthermore, the laptop’s battery was removed to ensure that the device under test was powered directly from the socket connected to the power meter.

For smartphones and tablets, the power meter could not be utilized as these devices do not function without a battery when plugged directly into the power outlet. Instead, PowerTutor was used to measure the power consumption of both devices. This tool has been utilized in various studies related to power measurement of components within Android-based devices [17, 18].

D. Experimention Procedures

To commence the experiment, the research team activated the power measurement tool for the designated device under examination. For the Raspberry Pi, desktop computer, and laptop, the device was initially turned off and then connected to the power meter, ensuring that the meter displayed a reading of 0.0W. In the case of the smartphone and tablet, PowerTutor was launched, and profiling commenced within the application. Once the power measurement mechanism was initiated, the first operation listed in Table VI was executed.

If the operation required internet connectivity, network access was enabled. Otherwise, Wi-Fi connectivity was disabled to prevent background processes from utilizing the network, thereby potentially increasing the device’s power consumption. Throughout the operation, readings from the power measurement tool were recorded at 2-second intervals for up to 2 minutes, ensuring an adequate number of values for reliability.

For instance, to measure the power consumption of O1, once the power button of the device under test was pressed, readings from the power meter were captured every 2 seconds until the welcome screen appeared. These recorded values were then averaged. This process was repeated three times to ensure result reliability, and the overall average for the three instances was calculated. The same procedure was replicated for all operations listed in Table VI and across the various devices under examination.

Upon conclusion of the experiment, the collected data were subjected to analysis using a statistical package, specifically SPSS.

4. Results and Discussions

The mean power readings obtained from the three iterations of the experiment with each device under evaluation are presented in Table VII.

Throughout the experiment, it was noted that both the electronic wattmeter and PowerTutor recorded power consumption at regular one-second intervals. Furthermore, according to Table VII, certain operations could not be finalized due to constraints elaborated in subsequent sections, as efforts were made to address RQ1 and RQ2.

A. R Pi Power Consumption Analysis(RQ1)

During the RPi experiment, three operations—O15 and O16—could not be conducted because no compatible version of Skype was available for installation on the RPi at the time of the experiment. Although Skype could potentially be installed through a virtual machine like ExaGear Desktop, doing so would introduce inconsistencies in the experimental setup and additional parameters to measure, such as the power consumption of the virtual machine. Additionally, O19 could not be measured as the RPi Model B does not have a sleep mode feature.

The results presented in Table VII indicate that each usage scenario consumes more power than when the device is in idle mode (O2). This suggests that both software usage and connected hardware have an impact on the overall power consumption of the Raspberry Pi. Furthermore, the power consumed by the RPi increased with the addition of more connected devices. For example, without any attached external hardware, the RPi’s power consumption was 2.10W, which increased to 2.20W when a mouse and keyboard were connected.

Additionally, the power consumption of each operation varied, with no two operations showing the same average power consumption. This variability can be attributed to various factors affecting the CPU’s power consumption during operations, including workload and processing speed. Overall, the average power consumption across the investigated operations was 2.84W, only 0.60W higher than idle mode power consumption. This relatively small difference highlights the RPi’s reasonably low overall power consumption during common operations. Among the operations investigated, watching an online video and playing an online game consumed the most power.

From the collected results, it was observed that operations requiring internet connectivity consumed relatively more power than those performed offline. For example, watching an online video consumed 4.00W compared to 2.80W for watching the same video via the default video player on the device under test. This disparity is due to the significant energy cost associated with wireless communication on computational devices. Overall, operations requiring internet connectivity averaged 3.68W compared to 2.58W for those not requiring internet connectivity. These trends are visually depicted in the boxplot shown in Fig. I.

B. Comparison of R Pi Power against Other Platforms(RQ2)

In comparison to laptops and desktop computers, several operations on smartphones and tablets could not be measured, specifically O1 (device start-up), O19 (sleep mode), and O20 (system shutdown). This limitation arose because PowerTutor was unable to operate during these modes, namely during device startup, shutdown, and while in sleep mode. Similar to the RPi, all other platforms tested exhibited minimal power consumption when the devices were in idle mode (O2). Additionally, operations requiring internet connectivity consumed more power than those performed offline. A line graph illustrating the power consumption comparisons among the various devices under test is provided in Fig II.

As illustrated in Fig. II, the various platforms under examination exhibited a similar power consumption trend for most operations investigated. However, a significant deviation was noted for O8—playing an installed game—where the laptop demonstrated a 157% increase in average power consumption compared to a 32% rise for the desktop computer and a 5% increase for the RPi, relative to idle power. Despite using the same game for both the laptop and desktop computer, the variance could be attributed to the differing specifications of the computers under test.

Overall, the results indicated that the desktop computer was the highest power consumer, averaging 39.96W, which is more than 10 times higher than the average power consumption of the RPi for the same set of operations. Notably, under sleep mode, the desktop computer continued to consume 29.20W, compared to only 1.20W for the laptop. While the laptop consumed less average power than the desktop computer at 17.68W, it still consumed over four times more power than the RPi. The smartphone and tablet emerged as the least power consumers of the experiment, averaging 0.44W and 0.33W, respectively, underscoring significant optimization efforts to reduce power consumption in battery-operated devices.

The study also encountered several challenges. Chief among them was the fluctuating values in power measurements for various operations, attributed to the software and hardware components within the devices under test. Operations requiring internet connectivity, such as web surfing (O5), playing a browser game (O6), and using messenger apps for calls (O15 and O16), exhibited the highest fluctuations, likely due to user-triggered events like voice and video inputs, mouse clicks, etc. To ensure reliability, power measurements were periodically recorded, and experiments were repeated multiple times. Standardizing operation procedures, such as consistent typing behavior for O13 and slide change intervals for O14, also helped ensure reliable power readings.

In terms of limitations, the inability to use a standardized power measurement mechanism for all devices was a major constraint. The experiment employed three different mechanisms—electronic wattmeter, PowerTutor, and Microsoft Joulemeter—and the inaccuracies of the software approach to measure power could affect recorded measurements. Additionally, the use of different tools resulted in varying precisions, with PowerTutor measuring in mW and the electronic wattmeter providing measurements in W (with one decimal place).

In conclusion, the experiment demonstrated that the RPi can be considered a low-power device compared to desktop computers and laptops. As a low-cost and low-power device, the RPi has the potential to save both money and energy if effectively integrated into households. However, adopting best practices can further reduce costs when using the RPi.

5. Further Reducing R Pi Power Consumption

To further diminish the power usage of the RPi and address RQ3, several methods and optimal strategies can be implemented during its functioning. These include:

1. Disconnect Peripherals that are not in use

Throughout the experiment, it became evident that adding additional peripherals to the RPi had an impact on its power consumption. One simple method to conserve power with the RPi is by deactivating or unplugging peripherals that are not actively being utilized. For example, if a connected printer or webcam is not currently in use with the RPi, it can be disconnected.

2.Work offline when ever possible