Скачать или смотреть 2025 Zkast

Zeus Kerravala from ZK Research is at NVIDIA's GTC event in DC to talk about the crucial role of networking in scaling AI. He's joined by WWill Eatherton, SVP of Network Engineering for Cisco, and Gilad Shainer, SVP of Networking for NVIDIA.

While GPUs get the glory, the network is the essential "super glue" that turns a collection of racks into a single, cohesive unit of computing—the AI Data Center. Will and Gilad discuss their decade-long partnership, the urgency of annual technology cadence in AI, and the key announcements driving the next generation of AI infrastructure.

🧠 Key Highlights & Announcements:
Networking's Critical Role in AI: The network's primary mission is maximizing GPU utilization for high-efficiency training and inference. It is the core component that allows the entire data center to function as a single, large supercomputer.

The NVIDIA AI Stack: Gilad explains why NVIDIA acquired Mellanox: to build the comprehensive infrastructure needed to connect all computing silicons. He differentiates between the high-performance InfiniBand for scale-up clusters and the specialized Spectrum-X Ethernet infrastructure designed specifically for distributed computing in AI, enabling broader adoption across enterprises and cloud providers.

The Cisco-NVIDIA Partnership: Will highlights that the partnership is driven by customer demand for Cisco's enterprise operational model, support, and optics, combined with NVIDIA's cutting-edge AI infrastructure. This collaboration delivers the best of both worlds, simplifying the management of complex AI supercomputers.

New Products & Innovation:

Cisco Nexus 9100: Announced as the first third-party switch supporting the NVIDIA Certified Professional (NCP) reference architecture for AI clusters, running both NX-OS and SONiC.

NVIDIA BlueField-4 DPU: A massive jump in performance, doubling the bandwidth, increasing compute capacity by 6x, and memory by 3x. The DPU remains the core device for securing the infrastructure and offloading storage processing in the north-south network.

Cisco Cloud Reference Architecture: Provides fully performance-validated, step-by-step guides for customers to build secure AI Factories, drastically cutting down deployment and optimization time from months to weeks.

Networking is Infrastructure: Gilad makes a strong case for moving beyond the term "networking," arguing that this technology is a crucial, compute-enabled computing infrastructure that does more than just move data.

Watch the full discussion to learn how these two leaders are ensuring the infrastructure can keep pace with the accelerating demands of AI!

👍 Like this video and hit the subscribe button for more ZKast deep dives!