Abstract: -This research introduces a comprehensive study of the most robust Single-Board Computers (SBCs)

implemented recently, where most of them are built on the system-on-chip architecture. This study also

presents the main characteristics of each of these SBCs, as well as their prices and applications. Additionally,

the study reviews some machine learning (ML) and deep learning (DL) techniques, exploring their

implementation on SBCs. Finally, it displays some software tools on how to implement DL and ML projects on

SBCs.

A single-board computer (SBC) is a new type of

computer system built on a single-circuit board. The

main components of an SBC include: a central

processing unit (CPU) to execute processes, a

graphical processing unit (GPU) to accelerate graph

processing, the Input/Output (I/O) general purpose

ports, and a memory (program and data). Moreover,

we can find image coding and different types of

Arduino and Raspberry Pi SBCs emerge as the most

prominent and widely used SBCs. However, upon

the separate examination and review of the Arduino

reviews the main features of the electronic hardware

and the libraries of AI and DL. Section 4 introduces

the best AI and DL single-board computers. Section

5 demonstrates the main tools and libraries for ML

and DL. Finally, Section 6 presents a discussion and

a conclusion of the paper.

When it comes to AI, the focus is on mastering the

Internet of things, [2], [3], [4], [5], [6], [7].

Meanwhile, DL operates without direct human

supervision and can analyze both unlabeled and

labeled data independently, without requiring third-

party support. DL finds applications in object

tracking and classification, image recognition, face

recognition, robot navigation, and large language

modeling tasks such as language translation and text

generation, [4], [7]. Even more with DL —a

transformation from text to image or image to voice

humans, such as customer service operations, lead

generation, quality control, and fraud detection. In

various domains, AI has demonstrated the ability to

outperform humans in the execution of some

specific tasks.

in today's technology landscape, experiencing rapid

growth. It is widely used in many important

applications, such as Google Maps, Google

Translate, Google Assistant, and Alexa. Many real-

wide usages, such as automatic tagging suggestions

for friends on social media sites. The speech

recognition process is based on converting spoken

tasks) and a neural processing unit (NPU) (a

processor used for deep learning tasks).

supported with machine learning and deep learning

libraries.

kit-learn, TensorFlow, PyTorch, OpenCV,

GPU4Vision, and OpenVIDIA. For generating 3D

computer-generated objects on a SBC, the open-

source OpenGL graphics library stands out as a

significant tool, [8]. This graphics library controls

graphic accelerators, which facilitate the processing

of Swift scenes. When engaging in parallel

computing with the SBCs covered in this article, it

Additionally, a key open-source library to consider

for SBC projects is OpenCV, [9], [10].

efficient in processing images and videos, utilizing

library tools such as DirectX, OpenCL, CUDA, and

OpenGL. Its interfaces, including Python, C++, and

C, contribute to its platform adaptability and

versatility. It also offers a comprehensive aid for

interacting with sensors and analyzing sensor data in

AI and DL applications. The incorporation of the

Cambridge, UK. It is probable that the Raspberry Pi

5 will have a processor that is more powerful in

performance than its predecessors. It ensures low

power consumption for applications that range from

IoT projects to desktop computing.

include, [13]: a 64-bit Arm Cortex-A76, with

512KB L2 cache and a 2MB shared L3 cache, a

CPU with 2.4GHz quad-core, a RAM of

LPDDR4X-4267 SDRAM (4GB or 8GB), a GPU

800MHz Video Core VII GPU supporting OpenGL

ES 3.1, Vulkan 1.2, and a Micro SD card slot, with

support for high-speed SDR104 mode for storage.

For communication, it has Dual band WiFi (2.4

micro-HDMI ports (up to 4Kp60 supported) HEVC

decoder and HDR support, 2 × 4 lane MIPI

camera/display transceivers (3x bandwidth

improvement), PCIe 2.0 x1 interface for quick

peripherals (it needs a different M.2 HAT or another

adapter), and a standard 40-pin Raspberry Pi header.

Raspberry Pi 5, equipped with 4 ARM cores and

running at 2.4 GHz, is considered an important

increase in processing power. The transition from

quick and accurate processing of AI. Figure 3

presents the Nvidia Jetson Nano SBC and its

application.

Fig. 3: Nvidia Jetson Nano SBC and its application

Power and size; with twenty-one TOPS and a size

up 75 mm. the system manages to do top

performance.

NVIDIA Volta GPU and 48 tensor cores

underscores its capabilities in handling complex AI

workloads, particularly those involving deep

learning and neural networks. The performance of

this SBC is improved, as its CPU comes with a 6-

core NVIDIA ARMv8.2 at 64 bits. Figure 4 shows

represents a powerful system-on-module (SOM)

that has these main specifications:

Xavier NX 16GB provides server-class

performance, which can deliver up to 14 TOPS at

10W, 21 TOPS at 15W, or 20W.

NX is implemented to be energy-saving during

power consumption, which makes it suitable for

edge computing applications. Versatility: It can be

Fig. 5: NVIDIA Jetson AGX Xavier

functional power of the Nvidia Jetson AGX Xavier

suggest that it may not be the best fit for beginners

or those exploring the field. Its capabilities are

is designed specifically for AI, ML, and DL

applications. Its computational power is exceptional.

Compared to traditional computing solutions, it

consumes only a fraction of the power to deliver up

to 32 TOPS of AI performance. Additionally, it

provides more memory space, which suits the AI

and DL workloads. This SBC provides many

connectivity options, such as Gigabit Ethernet,

PCIe, HDMI, USB 3.1, and PCIe, [18].

in ML and DL workloads. In addition, there is an

HDMI 2.0 display, 32GB of eMMC 5.1 storage, and

an extra storage extension, the UFS/uSD Card

Socket.

Coral Dev Board Micro is a microcontroller board

that can run TensorFlow Lite models with

acceleration on the Coral Edge TPU. With its

onboard camera and microphone, this module

highlighted for its functional design and

Certainly, the following paragraphs present an

overview of some of the most effective and widely

Single-board computers (SBCs) have shown up as

vital models for many applications, such as

these compact, powerful SBCs. Moreover, SBCs

presented effective and high-performance

computing solutions and ideal platforms for AI

workloads. The availability of different operating

systems, such as Android, and Linux distributions,

and custom-built environments, improves the

versatility and flexibility of SBCs for AI

development.

that are advised to be used in AI and DL projects

Board and NVIDIA Jetson Xavier are prepared for

immediate AI implementation right after purchasing

them. In contrast, additional development SBCs,

like the Raspberry Pi 4 or the most up-to-date Pi 5,

may require add-ons and extra components such as

the Google Coral TPU Accelerator Intel or the

Neural Compute Stick to perform machine learning

or deep learning tasks efficiently, especially for

complicated projects that request extra

computational power.

AI projects. Advanced SBCs, such as BeagleBone

AI and HiKey 970, are recommended for expert

users in AI.