That’s a wrap from Cisco Live Amsterdam! What an amazing week of exciting announcements and innovations demonstrating the best of AMD and Cisco’s collaboration. It was awesome to connect with our joint customers, strategic partners, and the Cisco Security and Networking leadership teams.
The explosion of Generative AI applications, coupled with the highly distributed nature of modern applications, is challenging enterprise network and security architectures like never before. The additional complexity of applications living outside of traditional data centers in hybrid or secondary cloud environments is driving customers to adopt a security posture in the data center where compromises or exploits are assumed. AI is massively transforming everything enterprises do in a data center, and the approach to traditi.... This new approach to data center security goes beyond the traditional perimeter method, mandating the inspection and validation of all application flows to restrict lateral movement of potential cybersecurity exploits and enforce secure guardrails for applications at all layers of the infrastructure—workload, virtual machine (VM), operating system, containers, and within the network itself.
Our collaboration with Cisco on the N9300 Smart Switches introduces a revolutionary innovation for services embedded into the network. The switches, embedded with AMD Pensando™ DPUs, combine networking and security functions in a single device, centrally managed through the AI-native Hypershield cloud controller. By integrating DPUs, we transform data center networking platforms into high-capacity, multifunctional devices that efficiently handle complex processing tasks and deliver advanced services. This new switch category embeds L4 zone-based autonomous segmentation and stateful firewalling directly within the fabric, with AMD Pensando DPUs accelerating Hypershield's enforcement. This reduces TCO by eliminating separate firewall appliances, enhances cybersecurity compliance by providing pervasive autonomous segmentation and firewalling at wire-rate across data centers and cloud edges, and simplifies network architectures.
While Hypershield is the first application to take advantage of the power of the AMD Pensando DPUs built with a P4 programmable engine in the N9300 smart switch, the platforms are future-ready and capable of more services than security. With the programmability of the AMD Pensando DPU, customers can maximize their infrastructure investments by - embedding further services in the network through future DPU P4 software upgrades, such as event-based telemetry, large-scale NAT, IPsec, and load-balancing.
Undoubtedly, DPUs have become the de facto standard in cloud and hyperscale data centers for, offloading and accelerating large-scale Software-Defined Networking (SDN), security, and storage functions for multi-tenant hosting infrastructure. The unique domain-specific AMD Pensando DPU architecture is purpose-built for large-scale compute, combining general-purpose cores with P4 fully programmable silicon, allowing the DPU software to be enhanced for new protocols and services without hardware changes. Deployed in some of the world’s largest hyperscale environments – including IBM Cloud, Microsoft Azure and Oracle Cloud – the programmability of the AMD Pensando DPUs is a key feature for long-term infrastructure investment protection, allowing the hardware to be re-programmed to meet the demands of new protocols and services.
At Cisco Live, we showcased another DPU-enabled hyperscale innovation with the Cisco 8102-DPU Smart Switch, which combines Cisco’s Silicon One Q200 ASIC with AMD Pensando DPUs and the open SONiC operating system. This platform decouples the SDN stack from the server smartNIC, allowing the DPUs in the smart switch to be utilized to deliver orders of magnitude higher networking performance for customers' virtual appliances hosted on Microsoft Azure.
Disaggregating SDN and offloading onto AMD Pensando DPUs has resulted, per Microsoft, in a 40x improvement in new connection establishment for virtual machine and appliance workloads as part of the Azure Accelerated Connections service. This elevates performance dramatically higher than previously achievable, with the flexibility to accelerate future services in software like load balancing. As the platform gets deployed more holistically across Azure, Microsoft estimates it will deliver approximately $100 million in TCO savings over its lifespan through reduced server footprint and appliance consolidation, software cost reduction, enabling what its executives have called "limitless networking" in Azure for customers.
Cisco Live Amsterdam 2025 has undoubtedly been a landmark event, showcasing the power of AMD's partnership with Cisco in revolutionizing data center networking and security. The introduction of the N9300 Series Smart Switches with embedded AMD Pensando DPUs and Cisco Hypershield represents a paradigm shift in network architecture, fusing networking and security functions into a single high-performance device. This innovation addresses the growing challenges posed by the distributed nature of modern applications, offering a solution that enhances security, reduces TCO, and simplifies network management. The showcased Cisco 8102-DPU smart switch, combining Silicon One Q200 ASIC with AMD-Pensando DPUs and SONiC OS, further demonstrates the potential of DPU-enabled hyperscale innovation, particularly in cloud environments like Microsoft Azure.
AMD AI Networking innovations extend beyond DPU technology, with the recent addition of the AMD Pensando Pollara 400 AI NIC announced Q4 2024. Designed for back-end networking, the AI NIC reduces job completion times, accelerating AI workloads by increasing the speed of GPU-to-GPU communication for large models, including AI training, Generative AI, research, and more. Built with the AMD P4 programmable engine, the Pollara 400 is the industry’s first AI NIC offering the ability for customers to upgrade their hardware with software and firmware upgrades to meet future services, and standards, including those set by the Ultra Ethernet Consortium (UEC). The Pollara 400 is set to transform Ethernet AI networking, leading to dramatic improvements in job completion times, better congestion handling, and more predictable forwarding for Ethernet-enabled backend AI networks.
It's clear that the AMD-Cisco collaboration is at the forefront of driving transformative changes in network infrastructure, setting new standards for performance, security, and efficiency in the rapidly evolving landscape of AI data centers, cloud, and enterprise computing.