Connecting the world and beyond

Hardware & Computing Power Benchmark

Focuses on the hardware equipment capabilities underlying artificial intelligence computing, covering performance testing of deep neural network processing chips, baseline measurement of AI multimedia cluster computing power, interconnection architectures of heterogeneous accelerators, and energy efficiency testing metrics for large-scale parallel training hardware.

4.1 Chips & Cluster Benchmarks

Formulates unified performance testing benchmarks and evaluation methods for AI chips (e.g., DNN accelerators) and cloud AI computing clusters when executing machine vision and NLP training and inference tasks.

ITU-T F.748.11: Metrics and evaluation methods for a deep neural network processor benchmark (2020)
Establishes an evaluation benchmark framework and reference model set for cloud and mobile deep neural network chip processors executing training and inference workloads.

ITU-T F.748.18: Metric and evaluation methods for AI-enabled multimedia application computing power benchmark (2022)
For AI computing systems supporting multimedia services, establishes application scenario computing power measurement baseline workloads and automated parameter configuration guidance.

F.748.18 Architecture framework of AI computing power benchmark

4.2 Interconnection & Energy Efficiency

Standardizes the interconnection network architecture protocols between multi-node AI accelerator hardware, as well as the architectural and energy efficiency requirements for large language model massive parallel training systems.

ITU-T F.AII: Technical Requirements of AI Accelerators Interconnection Framework for Multimedia Application (2025)
Standardizes the network, adaptation, and security management requirements for high-speed interconnected communication architectures between AI accelerator hardware for multimedia computing tasks.

ITU-T TR.FRES: Power Efficiency metrics and Evaluation Requirements for Foundation Model Training and Inference Systems (2025)
Technical report. Focuses on measuring the power consumption and energy efficiency of foundation model hardware training and inference systems, proposing "Green AI" evaluation metrics like throughput per watt.

ITU-T TR.FRMPT: Framework for large language model parallel training system (2025)
Technical report. Describes the parallel training strategy framework, memory partitioning, and cluster monitoring management requirements for large language models on GPU clusters.

Connecting the world and beyond

Hardware & Computing Power Benchmark

4.1 Chips & Cluster Benchmarks

​

4.2 Interconnection & Energy Efficiency