Year: 2024
Type: article
Abstract: AI hardware gas pedal is a hardware device specialized in accelerating the processing of Artificial Intelligence (AI) tasks, and its development history is closely related to the development of AI field. This will explore the theoretical foundation, development history, structural principles, and characteristics of AI hardware gas pedals. The theoretical foundations stem from the significant demands of deep learning algorithms for large-scale matrix operations, where traditional CPUs and GPUs encounter performance limitations. The development history of these accelerators shows a progression from general-purpose processors to the adoption of GPUs, which initially offered substantial improvements. However, as neural network models grew in complexity, GPUs too began to show bottlenecks. This led to the creation of specialized AI hardware accelerators which, through continual innovation, have become integral to AI computing. These devices are characterized by their efficient performance, low power consumption, and adaptability, offering a higher energy-efficiency ratio and faster processing for tasks like inference and training compared to conventional processors. Moreover, their architecture is optimized with customized computing cores, specialized storage systems, and effective data pathways to enhance the handling of extensive neural network models.
DOI: 10.54097/1b96k506
Author: Jiaming Hu
Institution: Xi’an Jiaotong-Liverpool University
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4407820526