In-Depth: Keysight AI

I'm Ram Periakaruppan, vice president and general manager of the network applications and security products at Keysight Technologies. At Keysight my role is focused on leading our next generation AI infrastructure validation along with our network testing and cybersecurity solutions. AI workloads are transforming data centers at an unprecedented pace and ensuring their performance, scalability and efficiency is a complex challenge.

Our team is at the forefront of AI R&D, developing advanced benchmarking, validation and optimization solutions that help the industry navigate the rapidly evolving landscape of AI powered infrastructure. I've had the privilege of working with some of the world's most innovative cloud and enterprise customers, helping them accelerate AI deployments, optimizing network performance and enhancing security. I'm excited to be here today to share more about how we're helping shape the future of AI driven data centers.

Let's dive in. Large scale AI training jobs often encounter bottlenecks, including power constraints, bandwidth inefficiencies, and component failures, each of which can severely impact the performance and scalability of AI data centers. Any inefficiencies in GPU utilization amplify these constraints. Sustained traffic implies massive data movement between GPUs, leading to possible network congestion pushing the limits of existing data center infrastructure. Components may pass every performance pressure test and still fail at the system level, meaning that a single weak link can have a system wide impact on AI data centers leading to increased job completion times.

The industry has recognized that this highly interconnected system demand full stack validation and optimization. There's a need to shift the validation to earlier phases in the design and manufacturing cycles. To help optimize the AI infrastructure, we're introducing our new architecture Keysight AI, otherwise known as KAI. The main foundational solution of this new architecture is the KAI Data Center Builder. KAI is a suite of solutions designed to address the challenge of early validation starting with pre-silicon leading to devices and components and finally system wide qualification. KAI architecture is focused on 4 areas, compute, interconnect, network and power, each with a powerful portfolio of solutions. Keysight covers the full breadth of solutions across these four areas, covering all layers of the technology stack from physical to application. Ultimately, reducing the risk of workload failures and increasing AI data center operating efficiency.

One of our customers, a Tier 1 cloud provider, uses the KAI Data Center Builder to Benchmark network design of their new AI training cluster. They needed to make sure that their 800 Gigabit Ethernet network is ready for build out and can support not only the current GPU model, but also the newer generation of GPUs which are not yet available. They were able to emulate the behavior of AI workloads specific to both types of systems and future proof the deployment by avoiding the optimization only for the current generation. As AI driven workloads continue to evolve, the demand for optimized, scalable, and high-performance AI infrastructure has never been greater. Power constraints, bandwidth inefficiencies and network level failures are challenges that AI data centers must overcome.

At Keysight we're tackling these challenges with our KAI portfolio of products and KAI Data Center Builder, which helped design, validate, optimize, emulate, and benchmark AI infrastructure at scale. We are excited to partner with AI accelerator vendors, network, and compute hardware providers, and hyperscale AI operators to push the boundaries of AI infrastructure. We look forward to working together to tackle the toughest challenges in AI to build the next generation of AI data centers.