Return
Intelligence Evaluating Computational Power: A Multi-Factor Method
DOI:10.1109/ACCESS.2025.3538977.png)
Abstract
En 中文
With the rapid development of hardware technologies and the continuous optimization of deep learning tasks, accurately evaluating the computational power of artificial intelligence (AI) systems has become a critical requirement. However, the diversity of hardware architectures and the complexity of deep learning tasks have hindered the establishment of a widely accepted unified evaluation standard. To address this issue, we propose a novel evaluation method that integrates key metrics such as model performance, computational throughput, and energy efficiency, providing a more comprehensive and reliable assessment of AI hardware computational capabilities. Through benchmark tests on servers, statistical analysis of the method, and case studies in cloud, edge, and terminal environments, this approach not only ensures the accuracy of existing evaluation frameworks but also reveals the key characteristics and potential influencing factors of hardware computational power from multiple dimensions.
Keywords:
Computational modeling
Benchmark testing
Artificial intelligence
Training
Inference algorithms
Deep learning
Analytical models
Performance evaluation
Accuracy
Predictive models
Deep learning benchmark test
AI hardware performance evaluation
model performance assessment
Journal
IF:
3.6
Papers:
9.8W
Citations:
29.4W
Organization
Cited Papers
The role of solvent in the self-assembly of m-aminobenzoic acid: a density functional theory and molecular dynamics study
CrystEngComm
IF0
Unravelling the biosynthesis of pyriculol in the rice blast fungus Magnaporthe oryzae
Microbiology
IF0

