Keysight introduces AI Data Center Builder to validate and optimize network architecture and host design

Keysight Technologies has launched the KAI Data Center Builder, a software suite designed to emulate real-world workloads for evaluating the performance of Artificial Intelligence (AI) training. The KAI Data Center Builder integrates training workloads of large language models with the design and validation of AI infrastructure, including network components and accelerators.

AI operators can utilize various parallel processing strategies to optimize training performance, wherein the KAI Data Center Builder supports experimentation with configuration settings and model partitioning. Key questions addressed during AI cluster design focus on data movement efficiency, including Graphics Processing Unit (GPU) interconnects and network load balancing.

The new solution enables customers to gain insights into performance bottlenecks and improve communication among AI training jobs. Keysight provides a library of workloads, allowing customers to test various model partitioning schemas effectively.

Random Access Memory (RAM) Periakaruppan, Vice President at Keysight, emphasized the importance of early validation in the design process to avoid costly rework. The KAI Data Center Builder aims to enhance the realism of AI workload emulation, optimally aligning components for effective system performance.

Keysight plans to showcase the KAI Data Center Builder at the OFC 2025 conference, highlighting its capabilities to aid AI operators and infrastructure vendors in validating designs and enhancing the efficiency of AI processing.