The Keysight AI (KAI) Data Centre Builder and the KAI architecture collectively provide a robust framework for validating and optimising AI infrastructure, ensuring that data centre (DC) operators can enhance system performance while managing the complexities inherent in modern data centres.
The KAI Data Centre Builder is an advanced software suite that emulates real-world workloads to assess how various algorithms, components, and protocols impact AI training performance. This capability allows AI operators to conduct thorough evaluations without the need for costly large-scale deployments.
By integrating workloads from large language models (LLMs) such as GPT and Llama, the KAI Data Centre Builder enables tighter interaction between hardware designs and AI training algorithms, ultimately boosting system efficacy.
A critical aspect of AI training is model partitioning, which involves various parallel processing strategies to expedite the training process. The KAI Data Centre Builder facilitates experimentation with parameters such as partition sizes and distribution. This is vital for understanding how communication patterns between graphics processing units (GPUs) affect overall job completion time. By simulating real-world AI training jobs, the software helps identify performance bottlenecks and optimise network utilisation, helping operators fine-tune their systems effectively.
"With AI infrastructure growing in scale and complexity, the necessity for full-stack validation and optimisation is paramount," said Ram Periakaruppan, vice president and general manager of Network Test & Security Solutions at Keysight. "The KAI Data Centre Builder brings realism to AI component design, ensuring that workloads are optimised for peak performance."
Complementing this software is the KAI architecture, a comprehensive portfolio of end-to-end solutions designed to scale AI processing capacity in data centres. This architecture addresses every facet of AI data centre design, from the physical layer through to the application layer, providing insights that enhance system-level interoperability and performance.
The KAI architecture comprises four key suites:
- KAI Data Centre Builder: Emulates high-scale AI workloads, improving system performance and optimising operations.
- KAI Compute: Focuses on high-speed digital designs and next-generation AI chip development.
- KAI Interconnect: Validates optical and electrical data paths to ensure high-speed connectivity.
- KAI Network: Benchmarks AI network performance and optimises workload distribution.
Each suite is equipped with AI-ready tools designed to meet the unique demands of AI infrastructure, ensuring that operators can validate and optimise their systems effectively. This holistic approach allows for early detection of potential issues, significantly reducing the risk of workload failures during production deployment.
Alan Weckel, founder and technology analyst at 650 Group, noted the importance of accelerating the design and deployment of AI/ML ASICs. "AI interconnect through scale-up, scale-out, and frontend networks will drive record 800GE and 1.6T port shipments over the next several years with one of the fastest innovation cycles to ever occur in the industry."
As organisations increasingly rely on AI technologies, Keysight’s solutions provide a much-needed pathway to maximise performance and efficiency in data centres. By enabling operators to validate their designs against real-world scenarios, Keysight ensures that businesses can confidently navigate the complexities of AI infrastructure while optimising their investments.
